chore(i18n): sync translations with latest source changes (chunk 65/148, 100 files)
This commit is contained in:
localizeflow[bot]
parent
8c519d5ea5
commit
29f3936eb1
|
|
@ -1,20 +1,20 @@
|
|||
<!--
|
||||
CO_OP_TRANSLATOR_METADATA:
|
||||
{
|
||||
"original_hash": "4ed9993bca581850c983c95d5a3f57eb",
|
||||
"translation_date": "2025-12-25T01:29:06+00:00",
|
||||
"original_hash": "14816e97d79b296c87811724f7785923",
|
||||
"translation_date": "2026-01-01T13:10:57+00:00",
|
||||
"source_file": "README.md",
|
||||
"language_code": "bg"
|
||||
}
|
||||
-->
|
||||
[](https://github.com/microsoft/AI-For-Beginners/blob/main/LICENSE)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/graphs/contributors/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/issues/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/issues/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/pulls/)
|
||||
[](http://makeapullrequest.com)
|
||||
[](http://makeapullrequest.com)
|
||||
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/watchers/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/network/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/watchers/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/network/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/stargazers/)
|
||||
[](https://mybinder.org/v2/gh/microsoft/ai-for-beginners/HEAD)
|
||||
[](https://gitter.im/Microsoft/ai-for-beginners?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge)
|
||||
|
|
@ -23,112 +23,112 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
# Изкуствен интелект за начинаещи - Учебна програма
|
||||
|
||||
||
|
||||
||
|
||||
|:---:|
|
||||
| AI For Beginners - _Скетчноут от [@girlie_mac](https://twitter.com/girlie_mac)_ |
|
||||
| Изкуствен интелект за начинаещи - _Sketchnote от [@girlie_mac](https://twitter.com/girlie_mac)_ |
|
||||
|
||||
Explore the world of **Изкуствен интелект** (ИИ) with our 12-week, 24-lesson curriculum! It includes practical lessons, quizzes, and labs. The curriculum is beginner-friendly and covers tools like TensorFlow and PyTorch, as well as ethics in AI
|
||||
Проучете света на **Изкуствения интелект** (ИИ) с нашата 12-седмична учебна програма от 24 урока! Тя включва практически уроци, тестове и лаборатории. Учебната програма е подходяща за начинаещи и обхваща инструменти като TensorFlow и PyTorch, както и етиката в ИИ
|
||||
|
||||
### 🌐 Поддръжка на много езици
|
||||
### 🌐 Поддръжка на множество езици
|
||||
|
||||
#### Поддържани чрез GitHub Action (автоматизирано и винаги актуално)
|
||||
#### Поддържа се чрез GitHub Action (Автоматично и винаги актуално)
|
||||
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE START -->
|
||||
[Арабски](../ar/README.md) | [Бенгалски](../bn/README.md) | [Български](./README.md) | [Бирмански (Мианмар)](../my/README.md) | [Китайски (опростен)](../zh/README.md) | [Китайски (традиционен, Хонг Конг)](../hk/README.md) | [Китайски (традиционен, Макао)](../mo/README.md) | [Китайски (традиционен, Тайван)](../tw/README.md) | [Хърватски](../hr/README.md) | [Чешки](../cs/README.md) | [Датски](../da/README.md) | [Нидерландски](../nl/README.md) | [Естонски](../et/README.md) | [Финландски](../fi/README.md) | [Френски](../fr/README.md) | [Немски](../de/README.md) | [Гръцки](../el/README.md) | [Иврит](../he/README.md) | [Хинди](../hi/README.md) | [Унгарски](../hu/README.md) | [Индонезийски](../id/README.md) | [Италиански](../it/README.md) | [Японски](../ja/README.md) | [Каннада](../kn/README.md) | [Корейски](../ko/README.md) | [Литовски](../lt/README.md) | [Малайски](../ms/README.md) | [Малаялам](../ml/README.md) | [Марати](../mr/README.md) | [Непали](../ne/README.md) | [Нигерийски пиджин](../pcm/README.md) | [Норвежки](../no/README.md) | [Персийски (фарси)](../fa/README.md) | [Полски](../pl/README.md) | [Португалски (Бразилия)](../br/README.md) | [Португалски (Португалия)](../pt/README.md) | [Пенджаби (Gurmukhi)](../pa/README.md) | [Румънски](../ro/README.md) | [Руски](../ru/README.md) | [Сръбски (кирилица)](../sr/README.md) | [Словашки](../sk/README.md) | [Словенски](../sl/README.md) | [Испански](../es/README.md) | [Суахили](../sw/README.md) | [Шведски](../sv/README.md) | [Тагалог (Филипински)](../tl/README.md) | [Тамилски](../ta/README.md) | [Телугу](../te/README.md) | [Тайландски](../th/README.md) | [Турски](../tr/README.md) | [Украински](../uk/README.md) | [Урду](../ur/README.md) | [Виетнамски](../vi/README.md)
|
||||
[арабски](../ar/README.md) | [бенгалски](../bn/README.md) | [български](./README.md) | [бирмански (Мианмар)](../my/README.md) | [Китайски (опростен)](../zh/README.md) | [Китайски (традиционен, Хонконг)](../hk/README.md) | [Китайски (традиционен, Макао)](../mo/README.md) | [Китайски (традиционен, Тайван)](../tw/README.md) | [хърватски](../hr/README.md) | [чешки](../cs/README.md) | [датски](../da/README.md) | [холандски](../nl/README.md) | [естонски](../et/README.md) | [финландски](../fi/README.md) | [френски](../fr/README.md) | [немски](../de/README.md) | [гръцки](../el/README.md) | [иврит](../he/README.md) | [хинди](../hi/README.md) | [унгарски](../hu/README.md) | [индонезийски](../id/README.md) | [италиански](../it/README.md) | [японски](../ja/README.md) | [каннада](../kn/README.md) | [корейски](../ko/README.md) | [литовски](../lt/README.md) | [малайски](../ms/README.md) | [малаялам](../ml/README.md) | [маратхи](../mr/README.md) | [непали](../ne/README.md) | [нигерийски пиджин](../pcm/README.md) | [норвежки](../no/README.md) | [персийски (фарси)](../fa/README.md) | [полски](../pl/README.md) | [португалски (Бразилия)](../br/README.md) | [португалски (Португалия)](../pt/README.md) | [панджаби (Гурумухи)](../pa/README.md) | [румънски](../ro/README.md) | [руски](../ru/README.md) | [сръбски (кирилица)](../sr/README.md) | [словашки](../sk/README.md) | [словенски](../sl/README.md) | [испански](../es/README.md) | [суахили](../sw/README.md) | [шведски](../sv/README.md) | [тагалог (филипински)](../tl/README.md) | [тамилски](../ta/README.md) | [телугу](../te/README.md) | [тайландски](../th/README.md) | [турски](../tr/README.md) | [украински](../uk/README.md) | [урду](../ur/README.md) | [виетнамски](../vi/README.md)
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE END -->
|
||||
|
||||
**Ако желаете да има допълнителни преводи, езиците, които се поддържат са изброени [тук](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md)**
|
||||
**Ако желаете да добавите допълнителни преводи, поддържаните езици са изброени [тук](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md)**
|
||||
|
||||
## Присъединете се към общността
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
## Какво ще научите
|
||||
|
||||
**[Мисловна карта на курса](http://soshnikov.com/courses/ai-for-beginners/mindmap.html)**
|
||||
**[Ментална карта на курса](http://soshnikov.com/courses/ai-for-beginners/mindmap.html)**
|
||||
|
||||
В тази учебна програма ще научите:
|
||||
|
||||
* Различни подходи към изкуствения интелект, включително "добрият стар" символичен подход с **Представяне на знания** и разсъждение ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence)).
|
||||
* **Невронни мрежи** и **Дълбоко обучение**, които са в основата на модерния ИИ. Ще илюстрираме концепциите зад тези важни теми с код в два от най-популярните фреймуърка - [TensorFlow](http://Tensorflow.org) и [PyTorch](http://pytorch.org).
|
||||
* **Невронни архитектури** за работа с изображения и текст. Ще разгледаме съвременни модели, но може да не обхващаме напълно най-новото състояние на техниката.
|
||||
* По-малко популярни подходи в ИИ, като **Генетични алгоритми** и **Мулти-агентни системи**.
|
||||
* Различни подходи към изкуствения интелект, включително „стария добър“ символен подход с **Представяне на знания** и умозаключение ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence)).
|
||||
* **Невронни мрежи** и **Дълбоко обучение**, които са в основата на съвременния ИИ. Ще илюстрираме концепциите зад тези важни теми чрез код в две от най-популярните рамки - [TensorFlow](http://Tensorflow.org) и [PyTorch](http://pytorch.org).
|
||||
* **Невронни архитектури** за работа с изображения и текст. Ще покрием по-нови модели, но може да сме малко по-дефицитни по отношение на най-новите постижения.
|
||||
* По-малко популярни подходи в ИИ, като **Генетични алгоритми** и **Многоагентни системи**.
|
||||
|
||||
Какво няма да обхванем в тази учебна програма:
|
||||
Какво няма да покрием в тази учебна програма:
|
||||
|
||||
> [Намерете всички допълнителни ресурси за този курс в нашата колекция Microsoft Learn](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)
|
||||
|
||||
* Бизнес случаи за използване на **ИИ в бизнеса**. Помислете да преминете обучителния път [Въведение в ИИ за бизнес потребители](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) в Microsoft Learn, или [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum), разработено в сътрудничество с [INSEAD](https://www.insead.edu/).
|
||||
* **Класическо машинно обучение**, което е добре описано в нашата [Учебна програма „Машинно обучение за начинаещи“](http://github.com/Microsoft/ML-for-Beginners).
|
||||
* Практически приложения на ИИ, изградени с използване на **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)**. За това препоръчваме да започнете с модулите в Microsoft Learn за [визия](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [обработка на естествен език](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Генеративен ИИ със Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** и други.
|
||||
* Конкретни ML **облачни рамки**, като [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum), или [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum). Помислете да използвате обучителните пътеки [Създаване и експлоатация на решения за машинно обучение с Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) и [Създаване и експлоатация на решения за машинно обучение с Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum).
|
||||
* **Разговорен ИИ** и **чатботове**. Съществува отделна обучителна пътека [Създаване на решения за разговорен ИИ](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum), и можете също да се позовете на [този блог пост](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/) за повече подробности.
|
||||
* **Дълбока математика** зад дълбокото обучение. За това бихме препоръчали [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) от Ian Goodfellow, Yoshua Bengio и Aaron Courville, който е наличен и онлайн на [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/).
|
||||
* Бизнес казуси за използване на **ИИ в бизнеса**. Помислете да преминете през учебния път [Въведение в ИИ за бизнес потребители](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) в Microsoft Learn, или [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum), разработен в сътрудничество с [INSEAD](https://www.insead.edu/).
|
||||
* **Класическо машинно обучение**, което е добре описано в нашата учебна програма [Machine Learning for Beginners Curriculum](http://github.com/Microsoft/ML-for-Beginners).
|
||||
* Практически приложения на ИИ, изградени с помощта на **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)**. За това препоръчваме да започнете с модули в Microsoft Learn за [компютърно зрение](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [обработка на естествен език](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Генеративен ИИ с Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** и други.
|
||||
* Специфични облачни рамки за машинно обучение, като [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum) или [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum). Помислете да използвате учебните пътеки [Build and operate machine learning solutions with Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) и [Build and Operate Machine Learning Solutions with Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum).
|
||||
* **Разговорен ИИ** и **чат ботове**. Има отделен учебен път [Create conversational AI solutions](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum), и можете също да се позовете на [този блог пост](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/) за повече детайли.
|
||||
* **Дълбока математика** зад дълбокото обучение. За това препоръчваме [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) от Ian Goodfellow, Yoshua Bengio и Aaron Courville, която е достъпна и онлайн на [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/).
|
||||
|
||||
За леко въведение в темите за _ИИ в облака_ може да помислите да преминете обучителната пътека [Започнете с изкуствен интелект в Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum) Learning Path.
|
||||
За меко въведение в темите за _ИИ в облака_ може да помислите да преминете учебния път [Започване с изкуствен интелект в Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum).
|
||||
|
||||
# Съдържание
|
||||
|
||||
| | Връзка към урока | PyTorch/Keras/TensorFlow | Лаборатория |
|
||||
| :-: | :------------------------------------------------------------------------------------------------------------------------------------------: | :---------------------------------------------------------------------------------------------: | ------------------------------------------------------------------------------ |
|
||||
| 0 | [Настройка на курса](./lessons/0-course-setup/setup.md) | [Настройване на среда за разработка](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| I | [**Въведение в ИИ**](./lessons/1-Intro/README.md) | | |
|
||||
| 01 | [Въведение и история на ИИ](./lessons/1-Intro/README.md) | - | - |
|
||||
| II | **Символичен ИИ** |
|
||||
| 02 | [Представяне на знания и експертни системи](./lessons/2-Symbolic/README.md) | [Експертни системи](./lessons/2-Symbolic/Animals.ipynb) / [Онтология](./lessons/2-Symbolic/FamilyOntology.ipynb) /[Граф на концепции](./lessons/2-Symbolic/MSConceptGraph.ipynb) | |
|
||||
| 0 | [Настройка на курса](./lessons/0-course-setup/setup.md) | [Настройка на вашата среда за разработка](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| I | [**Въведение в изкуствения интелект**](./lessons/1-Intro/README.md) | | |
|
||||
| 01 | [Въведение и история на изкуствения интелект](./lessons/1-Intro/README.md) | - | - |
|
||||
| II | **Символен ИИ** |
|
||||
| 02 | [Представяне на знания и експертни системи](./lessons/2-Symbolic/README.md) | [Експертни системи](./lessons/2-Symbolic/Animals.ipynb) / [Онтология](./lessons/2-Symbolic/FamilyOntology.ipynb) /[Концептуален граф](./lessons/2-Symbolic/MSConceptGraph.ipynb) | |
|
||||
| III | [**Въведение в невронните мрежи**](./lessons/3-NeuralNetworks/README.md) |||
|
||||
| 03 | [Перцептрон](./lessons/3-NeuralNetworks/03-Perceptron/README.md) | [Бележник](./lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb) | [Лаб](./lessons/3-NeuralNetworks/03-Perceptron/lab/README.md) |
|
||||
| 04 | [Многослоен перцептрон и създаване на собствен фреймуърк](./lessons/3-NeuralNetworks/04-OwnFramework/README.md) | [Бележник](./lessons/3-NeuralNetworks/04-OwnFramework/OwnFramework.ipynb) | [Лаб](./lessons/3-NeuralNetworks/04-OwnFramework/lab/README.md) |
|
||||
| 05 | [Въведение във фреймуърците (PyTorch/TensorFlow) и пренапасване](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Лаб](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**Компютърно зрение**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [Разгледайте компютърното зрение в Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 05 | [Въведение във фреймуъркове (PyTorch/TensorFlow) и overfitting](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Лаб](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**Компютърно зрение**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [Разгледайте компютърно зрение в Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 06 | [Въведение в компютърното зрение. OpenCV](./lessons/4-ComputerVision/06-IntroCV/README.md) | [Бележник](./lessons/4-ComputerVision/06-IntroCV/OpenCV.ipynb) | [Лаб](./lessons/4-ComputerVision/06-IntroCV/lab/README.md) |
|
||||
| 07 | [Конволюционни невронни мрежи](./lessons/4-ComputerVision/07-ConvNets/README.md) & [Архитектури на CNN](./lessons/4-ComputerVision/07-ConvNets/CNN_Architectures.md) | [PyTorch](./lessons/4-ComputerVision/07-ConvNets/ConvNetsPyTorch.ipynb) /[TensorFlow](./lessons/4-ComputerVision/07-ConvNets/ConvNetsTF.ipynb) | [Лаб](./lessons/4-ComputerVision/07-ConvNets/lab/README.md) |
|
||||
| 08 | [Предварително обучени мрежи и трансферно обучение](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [Трикове при обучение](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Лаб](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 08 | [Предварително обучени мрежи и трансферно обучение](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [Трикове при трениране](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Лаб](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 09 | [Автоенкодери и VAE](./lessons/4-ComputerVision/09-Autoencoders/README.md) | [PyTorch](./lessons/4-ComputerVision/09-Autoencoders/AutoEncodersPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/09-Autoencoders/AutoencodersTF.ipynb) | |
|
||||
| 10 | [Генеративни състезателни мрежи и пренос на художествен стил](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 10 | [Генеративни състезателни мрежи и артистичен трансфер на стил](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 11 | [Откриване на обекти](./lessons/4-ComputerVision/11-ObjectDetection/README.md) | [TensorFlow](./lessons/4-ComputerVision/11-ObjectDetection/ObjectDetection.ipynb) | [Лаб](./lessons/4-ComputerVision/11-ObjectDetection/lab/README.md) |
|
||||
| 12 | [Семантично сегментиране. U-Net](./lessons/4-ComputerVision/12-Segmentation/README.md) | [PyTorch](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationPytorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationTF.ipynb) | |
|
||||
| V | [**Обработка на естествен език**](./lessons/5-NLP/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-pytorch/?WT.mc_id=academic-77998-cacaste) /[TensorFlow](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-TensorFlow/?WT.mc_id=academic-77998-cacaste) | [Разгледайте обработката на естествен език в Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)|
|
||||
| V | [**Обработка на естествен език**](./lessons/5-NLP/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-pytorch/?WT.mc_id=academic-77998-cacaste) /[TensorFlow](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-TensorFlow/?WT.mc_id=academic-77998-cacaste) | [Разгледайте обработка на естествен език в Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)|
|
||||
| 13 | [Представяне на текст. BoW/TF-IDF](./lessons/5-NLP/13-TextRep/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationTF.ipynb) | |
|
||||
| 14 | [Семантични вграждания на думи. Word2Vec и GloVe](./lessons/5-NLP/14-Embeddings/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsTF.ipynb) | |
|
||||
| 15 | [Моделиране на езика. Обучение на собствени вграждания](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [Лаб](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| 15 | [Езиково моделиране. Трениране на собствени вграждания](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [Лаб](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| 16 | [Рекурентни невронни мрежи](./lessons/5-NLP/16-RNN/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNTF.ipynb) | |
|
||||
| 17 | [Генеративни рекурентни мрежи](./lessons/5-NLP/17-GenerativeNetworks/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativePyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativeTF.ipynb) | [Лаб](./lessons/5-NLP/17-GenerativeNetworks/lab/README.md) |
|
||||
| 18 | [Трансформъри. BERT.](./lessons/5-NLP/18-Transformers/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersPyTorch.ipynb) /[TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersTF.ipynb) | |
|
||||
| 19 | [Разпознаване на именовани същности](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [Лаб](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [Големи езикови модели, програмиране на подсказки и задачи с малко примери](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| VI | **Други техники в ИИ** || |
|
||||
| 19 | [Разпознаване на именовани обекти](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [Лаб](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [Големи езикови модели, prompt programming и задачи с малко примери](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| VI | **Други техники за ИИ** || |
|
||||
| 21 | [Генетични алгоритми](./lessons/6-Other/21-GeneticAlgorithms/README.md) | [Бележник](./lessons/6-Other/21-GeneticAlgorithms/Genetic.ipynb) | |
|
||||
| 22 | [Дълбоко подсилващо обучение](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [Лаб](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 23 | [Многоагентни системи](./lessons/6-Other/23-MultiagentSystems/README.md) | | |
|
||||
| 22 | [Дълбоко обучение с подсилване](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [Лаб](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 23 | [Системи с множество агенти](./lessons/6-Other/23-MultiagentSystems/README.md) | | |
|
||||
| VII | **Етика на ИИ** | | |
|
||||
| 24 | [Етика на ИИ и отговорен ИИ](./lessons/7-Ethics/README.md) | [Microsoft Learn: Принципи за отговорен ИИ](https://docs.microsoft.com/learn/paths/responsible-ai-business-principles/?WT.mc_id=academic-77998-cacaste) | |
|
||||
| IX | **Допълнително** | | |
|
||||
| IX | **Допълнения** | | |
|
||||
| 25 | [Мултимодални мрежи, CLIP и VQGAN](./lessons/X-Extras/X1-MultiModal/README.md) | [Бележник](./lessons/X-Extras/X1-MultiModal/Clip.ipynb) | |
|
||||
|
||||
## Всяка лекция съдържа
|
||||
## Всеки урок съдържа
|
||||
|
||||
* Материал за предварително четене
|
||||
* Изпълними Jupyter бележници, които често са специфични за фреймуърка (**PyTorch** или **TensorFlow**). Изпълнимият бележник съдържа също много теоретичен материал, така че за да разберете темата трябва да преминете през поне една версия на бележника (или PyTorch, или TensorFlow).
|
||||
* **Лабове** налични за някои теми, които ви дават възможност да опитате да приложите научения материал към конкретен проблем.
|
||||
* Някои секции съдържат връзки към [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) модули, които обхващат свързани теми.
|
||||
* Изпълними Jupyter бележници, които често са специфични за фреймуърка (**PyTorch** или **TensorFlow**). Изпълнимият бележник също съдържа много теоретичен материал, така че за да разберете темата трябва да преминете поне през една версия на бележника (или PyTorch, или TensorFlow).
|
||||
* **Лаборатории** налични за някои теми, които ви дават възможност да опитате да приложите наученото към конкретен проблем.
|
||||
* Някои секции съдържат връзки към [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) модули, които покриват свързани теми.
|
||||
|
||||
## Първи стъпки
|
||||
|
||||
### 🎯 Нов в ИИ? Започнете тук!
|
||||
|
||||
Ако сте напълно нови в ИИ и искате бързи, практични примери, разгледайте нашите [**Примери за начинаещи**](./examples/README.md)! Те включват:
|
||||
Ако сте напълно нов в областта на ИИ и искате бързи практически примери, разгледайте нашите [**Примери за начинаещи**](./examples/README.md)! Те включват:
|
||||
|
||||
- 🌟 **Здравей, AI свят** - Вашата първа програма за ИИ (разпознаване на образци)
|
||||
- 🧠 **Проста невронна мрежа** - Постройте невронна мрежа от нулата
|
||||
- 🖼️ **Класификатор на изображения** - Класифицирайте изображения с подробни коментари
|
||||
- 💬 **Тон на текста** - Анализира положителен/отрицателен текст
|
||||
- 🌟 **Здравей, свят на ИИ** - Вашата първа програма за ИИ (разпознаване на образци)
|
||||
- 🧠 **Проста невронна мрежа** - Изградете невронна мрежа от нулата
|
||||
- 🖼️ **Класификатор на изображения** - Класира изображения с подробни коментари
|
||||
- 💬 **Text Sentiment** - Analyze positive/negative text
|
||||
|
||||
Тези примери са създадени, за да ви помогнат да разберете концепциите за ИИ, преди да се потопите в пълния учебен план.
|
||||
These examples are designed to help you understand AI concepts before diving into the full curriculum.
|
||||
|
||||
### 📚 Пълна настройка на учебната програма
|
||||
### 📚 Full Curriculum Setup
|
||||
|
||||
- Създадохме a [setup lesson](./lessons/0-course-setup/setup.md), който да ви помогне да настроите вашата развойна среда. - За преподаватели, също сме създали a [curricula setup lesson](./lessons/0-course-setup/for-teachers.md) за вас!
|
||||
- Как да [Run the code in a VSCode or a Codepace](./lessons/0-course-setup/how-to-run.md)
|
||||
- We have created a [setup lesson](./lessons/0-course-setup/setup.md) to help you with setting up your development environment. - For Educators, we have created a [curricula setup lesson](./lessons/0-course-setup/for-teachers.md) for you too!
|
||||
- How to [Run the code in a VSCode or a Codepace](./lessons/0-course-setup/how-to-run.md)
|
||||
|
||||
Follow these steps:
|
||||
|
||||
|
|
@ -138,85 +138,85 @@ Clone the Repository: `git clone https://github.com/microsoft/AI-For-Beginners.g
|
|||
|
||||
Don't forget to star (🌟) this repo to find it easier later.
|
||||
|
||||
## Срещнете други обучаеми
|
||||
## Meet other Learners
|
||||
|
||||
Присъединете се към нашия [официален AI Discord сървър](https://aka.ms/genai-discord?WT.mc_id=academic-105485-bethanycheum) за да се срещнете и общувате с други обучаеми, които посещават този курс, и да получите подкрепа.
|
||||
Join our [official AI Discord server](https://aka.ms/genai-discord?WT.mc_id=academic-105485-bethanycheum) to meet and network with other learners taking this course and get support.
|
||||
|
||||
Ако имате обратна връзка за продукта или въпроси по време на разработката, посетете нашия [Azure AI Foundry Developer Forum](https://aka.ms/foundry/forum)
|
||||
If you have product feedback or questions whilst building visit our [Azure AI Foundry Developer Forum](https://aka.ms/foundry/forum)
|
||||
|
||||
## Тестове
|
||||
## Quizzes
|
||||
|
||||
> **Бележка относно тестовете**: Всички тестове са в папката Quiz-app в etc\quiz-app, или [Online Here](https://ff-quizzes.netlify.app/) Те са свързани от уроците; приложението за тестове може да се стартира локално или да бъде разгърнато в Azure; следвайте инструкциите в папката `quiz-app`. Те постепенно се локализират.
|
||||
> **A note about quizzes**: All quizzes are contained in the Quiz-app folder in etc\quiz-app, or [Online Here](https://ff-quizzes.netlify.app/) They are linked from within the lessons the quiz app can be run locally or deployed to Azure; follow the instruction in the `quiz-app` folder. They are gradually being localized.
|
||||
|
||||
## Търсим помощ
|
||||
## Help Wanted
|
||||
|
||||
Имате ли предложения или сте открили правописни или кодови грешки? Raise an issue or create a pull request.
|
||||
Do you have suggestions or found spelling or code errors? Raise an issue or create a pull request.
|
||||
|
||||
## Специални благодарности
|
||||
## Special Thanks
|
||||
|
||||
* **✍️ Главен автор:** [Dmitry Soshnikov](http://soshnikov.com), PhD
|
||||
* **🔥 Редактор:** [Jen Looper](https://twitter.com/jenlooper), PhD
|
||||
* **🎨 Илюстратор на скици:** [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
* **✅ Създател на тестове:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 Основни сътрудници:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
* **✍️ Primary Author:** [Dmitry Soshnikov](http://soshnikov.com), PhD
|
||||
* **🔥 Editor:** [Jen Looper](https://twitter.com/jenlooper), PhD
|
||||
* **🎨 Sketchnote illustrator:** [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
* **✅ Quiz Creator:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 Core Contributors:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
|
||||
## Други учебни програми
|
||||
## Other Curricula
|
||||
|
||||
Нашият екип подготвя и други учебни програми! Вижте:
|
||||
Our team produces other curricula! Check out:
|
||||
|
||||
<!-- CO-OP TRANSLATOR OTHER COURSES START -->
|
||||
### LangChain
|
||||
[](https://aka.ms/langchain4j-for-beginners)
|
||||
[](https://aka.ms/langchainjs-for-beginners?WT.mc_id=m365-94501-dwahlin)
|
||||
[](https://aka.ms/langchain4j-for-beginners)
|
||||
[](https://aka.ms/langchainjs-for-beginners?WT.mc_id=m365-94501-dwahlin)
|
||||
|
||||
---
|
||||
|
||||
### Azure / Edge / MCP / Агенти
|
||||
[](https://github.com/microsoft/AZD-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/edgeai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mcp-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/ai-agents-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
### Azure / Edge / MCP / Agents
|
||||
[](https://github.com/microsoft/AZD-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/edgeai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mcp-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/ai-agents-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Серия за генеративен ИИ
|
||||
[](https://github.com/microsoft/generative-ai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[-9333EA?style=for-the-badge&labelColor=E5E7EB&color=9333EA)](https://github.com/microsoft/Generative-AI-for-beginners-dotnet?WT.mc_id=academic-105485-koreyst)
|
||||
[-C084FC?style=for-the-badge&labelColor=E5E7EB&color=C084FC)](https://github.com/microsoft/generative-ai-for-beginners-java?WT.mc_id=academic-105485-koreyst)
|
||||
[-E879F9?style=for-the-badge&labelColor=E5E7EB&color=E879F9)](https://github.com/microsoft/generative-ai-with-javascript?WT.mc_id=academic-105485-koreyst)
|
||||
### Generative AI Series
|
||||
[](https://github.com/microsoft/generative-ai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[-9333EA?style=for-the-badge&labelColor=E5E7EB&color=9333EA)](https://github.com/microsoft/Generative-AI-for-beginners-dotnet?WT.mc_id=academic-105485-koreyst)
|
||||
[-C084FC?style=for-the-badge&labelColor=E5E7EB&color=C084FC)](https://github.com/microsoft/generative-ai-for-beginners-java?WT.mc_id=academic-105485-koreyst)
|
||||
[-E879F9?style=for-the-badge&labelColor=E5E7EB&color=E879F9)](https://github.com/microsoft/generative-ai-with-javascript?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Основно обучение
|
||||
[](https://aka.ms/ml-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/datascience-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ai-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/Security-101?WT.mc_id=academic-96948-sayoung)
|
||||
[](https://aka.ms/webdev-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/iot-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/xr-development-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
### Core Learning
|
||||
[](https://aka.ms/ml-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/datascience-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ai-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/Security-101?WT.mc_id=academic-96948-sayoung)
|
||||
[](https://aka.ms/webdev-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/iot-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/xr-development-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Серия Copilot
|
||||
[](https://aka.ms/GitHubCopilotAI?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mastering-github-copilot-for-dotnet-csharp-developers?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/CopilotAdventures?WT.mc_id=academic-105485-koreyst)
|
||||
### Copilot Series
|
||||
[](https://aka.ms/GitHubCopilotAI?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mastering-github-copilot-for-dotnet-csharp-developers?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/CopilotAdventures?WT.mc_id=academic-105485-koreyst)
|
||||
<!-- CO-OP TRANSLATOR OTHER COURSES END -->
|
||||
|
||||
## Получаване на помощ
|
||||
## Getting Help
|
||||
|
||||
Ако се затрудните или имате въпроси относно изграждането на AI приложения, присъединете се към други обучаеми и опитни разработчици в дискусии за MCP. Това е подкрепяща общност, където въпросите са добре дошли и знанието се споделя свободно.
|
||||
If you get stuck or have any questions about building AI apps. Join fellow learners and experienced developers in discussions about MCP. It's a supportive community where questions are welcome and knowledge is shared freely.
|
||||
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
Ако имате обратна връзка за продукта или грешки по време на разработка, посетете:
|
||||
If you have product feedback or errors while building visit:
|
||||
|
||||
[](https://aka.ms/foundry/forum)
|
||||
[](https://aka.ms/foundry/forum)
|
||||
|
||||
---
|
||||
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER START -->
|
||||
Отказ от отговорност:
|
||||
Този документ е преведен с помощта на услуга за превод с изкуствен интелект [Co-op Translator](https://github.com/Azure/co-op-translator). Въпреки че се стремим към точност, имайте предвид, че автоматизираните преводи могат да съдържат грешки или неточности. Оригиналният документ на оригиналния си език трябва да се счита за авторитетен източник. За критична информация се препоръчва професионален човешки превод. Ние не носим отговорност за каквито и да е недоразумения или неправилни тълкувания, произтичащи от използването на този превод.
|
||||
**Отказ от отговорност**:
|
||||
Този документ е преведен с помощта на AI услуга за превод [Co-op Translator](https://github.com/Azure/co-op-translator). Въпреки че се стремим към точност, моля, имайте предвид, че автоматичните преводи могат да съдържат грешки или неточности. Оригиналният документ на оригиналния език трябва да се счита за авторитетен източник. За критична информация се препоръчва професионален превод, извършен от човек. Ние не носим отговорност за никакви недоразумения или погрешни тълкувания, произтичащи от използването на този превод.
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER END -->
|
||||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Въведение в AI
|
||||
|
||||
|
||||
|
||||
|
||||
> Рисунка от [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Първоначално компютрите са били изобретени от [Чарлз Бабидж](https://en.wikipedia.org/wiki/Charles_Babbage), за да работят с числа, следвайки добре дефинирана процедура – алгоритъм. Съвременните компютри, макар и значително по-усъвършенствани от оригиналния модел, предложен през 19-ти век, все още следват същата идея за контролирани изчисления. Следователно е възможно да програмираме компютър да извърши нещо, ако знаем точната последователност от стъпки, които трябва да изпълним, за да постигнем целта.
|
||||
|
||||
|
||||
|
||||
|
||||
> Снимка от [Vickie Soshnikova](http://twitter.com/vickievalerie)
|
||||
|
||||
|
|
@ -46,7 +46,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Един от проблемите при работа с термина **[Интелигентност](https://en.wikipedia.org/wiki/Intelligence)** е, че няма ясно определение за този термин. Може да се твърди, че интелигентността е свързана с **абстрактно мислене** или със **самосъзнание**, но не можем да я дефинираме правилно.
|
||||
|
||||
|
||||
|
||||
|
||||
> [Снимка](https://unsplash.com/photos/75715CVEJhI) от [Amber Kipp](https://unsplash.com/@sadmax) от Unsplash
|
||||
|
||||
|
|
@ -98,13 +98,13 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
> | А какво да кажем за ML? | |
|
||||
> |--------------|-----------|
|
||||
> | Част от Изкуствения интелект, която се основава на компютърното обучение за решаване на проблем въз основа на някои данни, се нарича **Машинно обучение**. Няма да разглеждаме класическото машинно обучение в този курс – насочваме ви към отделната учебна програма [Машинно обучение за начинаещи](http://aka.ms/ml-beginners). |  |
|
||||
> | Част от Изкуствения интелект, която се основава на компютърното обучение за решаване на проблем въз основа на някои данни, се нарича **Машинно обучение**. Няма да разглеждаме класическото машинно обучение в този курс – насочваме ви към отделната учебна програма [Машинно обучение за начинаещи](http://aka.ms/ml-beginners). |  |
|
||||
|
||||
## Кратка история на AI
|
||||
|
||||
Изкуственият интелект започва като област в средата на двадесети век. Първоначално символичното разсъждение е преобладаващ подход и води до редица важни успехи, като експертни системи – компютърни програми, които могат да действат като експерт в някои ограничени проблемни области. Въпреки това скоро става ясно, че такъв подход не се мащабира добре. Извличането на знания от експерт, представянето им в компютър и поддържането на точна база от знания се оказва много сложна задача и твърде скъпа, за да бъде практична в много случаи. Това води до така наречената [AI зима](https://en.wikipedia.org/wiki/AI_winter) през 70-те години.
|
||||
|
||||
<img alt="Кратка история на AI" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a0264357672b0884cf3a220fbafe35c65d70b2c3805f7bf5e.bg.png" width="70%"/>
|
||||
<img alt="Кратка история на AI" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a.bg.png" width="70%"/>
|
||||
|
||||
> Изображение от [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
|
|||
|
|
@ -12,7 +12,7 @@
|
|||
"\n",
|
||||
"В този пример ще изградим проста система, базирана на знания, за определяне на животно въз основа на някои физически характеристики. Системата може да бъде представена чрез следното AND-OR дърво (това е част от цялото дърво, лесно можем да добавим още правила):\n",
|
||||
"\n",
|
||||
"\n"
|
||||
"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Представяне на знания и експертни системи
|
||||
|
||||
|
||||
|
||||
|
||||
> Скица от [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
@ -41,7 +41,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Следователно, проблемът с **представянето на знания** е да се намери ефективен начин за представяне на знанията вътре в компютър под формата на данни, за да бъдат автоматично използваеми. Това може да се разглежда като спектър:
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
@ -94,7 +94,7 @@ Python | синтаксис на блокове | отстъп
|
|||
|
||||
Един от ранните успехи на символния AI бяха така наречените **експертни системи** - компютърни системи, които бяха проектирани да действат като експерт в някаква ограничена проблемна област. Те се основаваха на **база знания**, извлечена от един или повече човешки експерти, и съдържаха **инференционен двигател**, който извършваше разсъждения върху нея.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
---------------------------------------------|------------------------------------------------
|
||||
Опростена структура на човешката нервна система | Архитектура на система, базирана на знания
|
||||
|
||||
|
|
@ -106,7 +106,7 @@ Python | синтаксис на блокове | отстъп
|
|||
|
||||
Като пример, нека разгледаме следната експертна система за определяне на животно въз основа на неговите физически характеристики:
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
|
|||
|
|
@ -1255,7 +1255,7 @@
|
|||
"* Ниска загуба при обучението - моделът може добре да приближи обучаващите данни, защото има достатъчно изразителна мощност.\n",
|
||||
"* Загубата при валидация може да бъде много по-висока от загубата при обучение и може да започне да се увеличава по време на обучението - това е, защото моделът \"запомня\" обучаващите точки и губи \"общата картина\".\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> На тази картинка, `x` представлява обучаващи данни, `o` - валидационни данни. Ляво - линеен модел (еднослоен), той приближава естеството на данните доста добре. Дясно - преобучен модел, моделът перфектно приближава обучаващите данни, но спира да има смисъл с всякакви други данни (грешката при валидация е много висока).\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ Overfitting е изключително важно понятие в машин
|
|||
|
||||
Разгледайте следния проблем за апроксимация на 5 точки (представени с `x` на графиките по-долу):
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-------------------------|--------------------------
|
||||
**Линеен модел, 2 параметъра** | **Нелинеен модел, 7 параметъра**
|
||||
Грешка при обучение = 5.3 | Грешка при обучение = 0
|
||||
|
|
@ -79,7 +79,7 @@ Overfitting е изключително важно понятие в машин
|
|||
|
||||
Както можете да видите от графиката по-горе, overfitting може да бъде открит чрез много ниска грешка при обучение и висока грешка при валидиране. Обикновено по време на обучение ще видим как грешките при обучение и валидиране започват да намаляват, но в даден момент грешката при валидиране може да спре да намалява и да започне да се увеличава. Това ще бъде знак за overfitting и индикатор, че вероятно трябва да спрем обучението на този етап (или поне да направим моментна снимка на модела).
|
||||
|
||||
|
||||
|
||||
|
||||
## Как да предотвратим overfitting
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Въведение в невронните мрежи
|
||||
|
||||
|
||||
|
||||
|
||||
Както обсъдихме във въведението, един от начините за постигане на интелигентност е чрез обучение на **компютърен модел** или **изкуствен мозък**. От средата на 20-ти век изследователите опитват различни математически модели, докато в последните години този подход не се оказа изключително успешен. Тези математически модели на мозъка се наричат **невронни мрежи**.
|
||||
|
||||
|
|
@ -36,13 +36,13 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
От биологията знаем, че нашият мозък се състои от нервни клетки (неврони), всяка от които има множество "входове" (дендрити) и един "изход" (аксон). Както дендритите, така и аксоните могат да провеждат електрически сигнали, а връзките между тях — известни като синапси — могат да имат различна степен на проводимост, която се регулира от невротрансмитери.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
Реален неврон *([Изображение](https://en.wikipedia.org/wiki/Synapse#/media/File:SynapseSchematic_lines.svg) от Wikipedia)* | Изкуствен неврон *(Изображение от автора)*
|
||||
|
||||
Следователно, най-простият математически модел на неврон съдържа няколко входа X<sub>1</sub>, ..., X<sub>N</sub> и един изход Y, както и серия от тегла W<sub>1</sub>, ..., W<sub>N</sub>. Изходът се изчислява като:
|
||||
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd767313e067719f400cec4b0e5090239c3e997c29f6789d4c3c263.bg.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd76731.bg.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
|
||||
където f е някаква нелинейна **активационна функция**.
|
||||
|
||||
|
|
|
|||
|
|
@ -73,14 +73,14 @@ im = cv2.cvtColor(im,cv2.COLOR_BGR2RGB)
|
|||
|
||||
* **Предварителна обработка на фотография на книга на Брайл**. Фокусираме се върху това как можем да използваме прагова обработка, откриване на характеристики, перспективна трансформация и манипулации с NumPy, за да отделим отделни символи на Брайл за по-нататъшна класификация от невронна мрежа.
|
||||
|
||||
 |  | 
|
||||
 |  | 
|
||||
----|-----|-----
|
||||
|
||||
> Изображение от [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
* **Откриване на движение във видео чрез разлика между кадри**. Ако камерата е фиксирана, тогава кадрите от камерата трябва да са доста подобни един на друг. Тъй като кадрите се представят като масиви, просто като извадите тези масиви за два последователни кадъра, ще получите разликата в пикселите, която трябва да е ниска за статични кадри и да стане по-висока, когато има значително движение в изображението.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
|
|
@ -89,7 +89,7 @@ im = cv2.cvtColor(im,cv2.COLOR_BGR2RGB)
|
|||
- **Плътен оптичен поток** изчислява векторно поле, което показва за всеки пиксел къде се движи.
|
||||
- **Рядък оптичен поток** се основава на вземане на някои отличителни характеристики в изображението (например ръбове) и изграждане на тяхната траектория от кадър на кадър.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
|
|
|
|||
|
|
@ -13,11 +13,11 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
VGG-16 е мрежа, която постигна 92.7% точност в класификацията на ImageNet топ-5 през 2014 г. Тя има следната структура на слоевете:
|
||||
|
||||
|
||||
|
||||
|
||||
Както можете да видите, VGG следва традиционна пирамидална архитектура, която представлява последователност от слоеве за конволюция и пулуване.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Researchgate](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493)
|
||||
|
||||
|
|
|
|||
|
|
@ -262,7 +262,7 @@
|
|||
"\n",
|
||||
"Така в типичен CNN ще има няколко слоя на свиване, със слоеве за пулуване между тях, за да се намалят размерите на изображението. Също така ще увеличим броя на филтрите, защото с напредването на моделите има повече възможни интересни комбинации, които трябва да търсим.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Поради намаляването на пространствените размери и увеличаването на размерите на характеристиките/филтрите, тази архитектура се нарича също **пирамидална архитектура**.\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -360,7 +360,7 @@
|
|||
"\n",
|
||||
"Така, в типичен CNN има няколко конволюционни слоя, със слоеве за пуллинг между тях, за да се намалят размерите на изображението. Също така увеличаваме броя на филтрите, защото с напредването на моделите има повече възможни интересни комбинации, които трябва да търсим.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Поради намаляването на пространствените размери и увеличаването на размерите на характеристиките/филтрите, тази архитектура се нарича **пирамидална архитектура**.\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -17,7 +17,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
За да извлечем шаблони, ще използваме понятието за **конволюционни филтри**. Както знаете, едно изображение се представя чрез 2D-матрица или 3D-тензор с цветова дълбочина. Прилагането на филтър означава, че вземаме сравнително малка матрица, наречена **ядро на филтъра**, и за всеки пиксел в оригиналното изображение изчисляваме претегленото средно с неговите съседни точки. Можем да си представим това като малък прозорец, който се плъзга по цялото изображение и осреднява всички пиксели според теглата в матрицата на ядрото на филтъра.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
|
||||
> Изображение от Дмитрий Сошников
|
||||
|
|
@ -38,7 +38,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
* Можем да проектираме мрежата така, че филтрите да се обучават автоматично
|
||||
* Можем да използваме същия подход, за да намираме шаблони в характеристики на високо ниво, а не само в оригиналното изображение. Така извличането на характеристики в CNN работи на йерархия от характеристики, започвайки от нискоуровневи комбинации от пиксели до по-високо ниво комбинации от части на изображението.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [статия на Hislop-Lynch](https://www.semanticscholar.org/paper/Computer-vision-based-pedestrian-trajectory-Hislop-Lynch/26e6f74853fc9bbb7487b06dc2cf095d36c9021d), базирано на [тяхното изследване](https://dl.acm.org/doi/abs/10.1145/1553374.1553453)
|
||||
|
||||
|
|
@ -55,9 +55,9 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Като пример, нека разгледаме архитектурата на VGG-16, мрежа, която постигна 92.7% точност в класификацията на ImageNet (топ 5) през 2014 г.:
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Researchgate](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493)
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Ще използваме [Oxford-IIIT Pet Dataset](https://www.robots.ox.ac.uk/~vgg/data/pets/), който съдържа изображения на 37 различни породи кучета и котки.
|
||||
|
||||
|
||||
|
||||
|
||||
За да изтеглите набора от данни, използвайте този кодов фрагмент:
|
||||
|
||||
|
|
|
|||
|
|
@ -50,7 +50,7 @@
|
|||
"\n",
|
||||
"За да визуализираме идеалната котка, ще започнем с изображение от случайни шумове и ще се опитаме да използваме техниката на оптимизация чрез градиентен спуск, за да коригираме изображението така, че мрежата да разпознае котка.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Ето нашето начално изображение:\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -29,7 +29,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Ето примерни характеристики, извлечени от снимка на котка от мрежата VGG-16:
|
||||
|
||||
|
||||
|
||||
|
||||
## Набор от данни за котки и кучета
|
||||
|
||||
|
|
@ -48,19 +48,19 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Един подход, който можем да използваме, е да започнем с произволно изображение и след това да използваме техниката **оптимизация чрез градиентен спуск**, за да го коригираме така, че мрежата да започне да мисли, че това е котка.
|
||||
|
||||
|
||||
|
||||
|
||||
Ако направим това, ще получим нещо много подобно на случаен шум. Това е така, защото *има много начини мрежата да мисли, че входното изображение е котка*, включително такива, които визуално не изглеждат логични. Докато тези изображения съдържат много модели, типични за котка, няма нищо, което да ги ограничава да бъдат визуално разпознаваеми.
|
||||
|
||||
За да подобрим резултата, можем да добавим друг термин към функцията за загуба, наречен **variation loss**. Това е метрика, която показва колко сходни са съседните пиксели на изображението. Минимизирането на variation loss прави изображението по-гладко и премахва шума, разкривайки по-визуално привлекателни модели. Ето пример за такива "идеални" изображения, които се класифицират като котка и като зебра с висока вероятност:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*Идеална котка* | *Идеална зебра*
|
||||
|
||||
Подобен подход може да се използва за извършване на така наречените **адверсариални атаки** върху невронна мрежа. Да предположим, че искаме да заблудим невронната мрежа и да направим така, че куче да изглежда като котка. Ако вземем изображение на куче, което мрежата разпознава като куче, можем да го коригираме малко чрез оптимизация с градиентен спуск, докато мрежата започне да го класифицира като котка:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*Оригинална снимка на куче* | *Снимка на куче, класифицирано като котка*
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@
|
|||
"\n",
|
||||
"Тъй като обучаваме автоенкодера да улавя възможно най-много информация от оригиналното изображение за точно възстановяване, мрежата се опитва да намери най-доброто **вграждане** на входните изображения, за да улови тяхното значение.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> Изображение от [Keras блог](https://blog.keras.io/building-autoencoders-in-keras.html)\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@
|
|||
"\n",
|
||||
"Тъй като тренираме автоенкодера да улавя колкото се може повече информация от оригиналното изображение за точно възстановяване, мрежата се опитва да намери най-доброто **вграждане** на входните изображения, за да улови тяхното значение.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Изображение от [Keras блог](https://blog.keras.io/building-autoencoders-in-keras.html)*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Тъй като обучаваме автоенкодера да улавя максимално много информация от оригиналното изображение за точна реконструкция, мрежата се опитва да намери най-доброто **вграждане** на входните изображения, за да улови тяхното значение.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Keras блог](https://blog.keras.io/building-autoencoders-in-keras.html)
|
||||
|
||||
|
|
|
|||
|
|
@ -13,7 +13,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
## [Тест преди лекцията](https://ff-quizzes.netlify.app/en/ai/quiz/21)
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [уебсайта на YOLO v2](https://pjreddie.com/darknet/yolov2/)
|
||||
|
||||
|
|
@ -25,7 +25,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
2. Извършваме класификация на изображения върху всяка част.
|
||||
3. Тези части, които водят до достатъчно висока активация, могат да се считат за съдържащи търсения обект.
|
||||
|
||||
|
||||
|
||||
|
||||
> *Изображение от [тетрадката с упражнения](ObjectDetection-TF.ipynb)*
|
||||
|
||||
|
|
@ -42,7 +42,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
* [PASCAL VOC](http://host.robots.ox.ac.uk/pascal/VOC/) - 20 класа
|
||||
* [COCO](http://cocodataset.org/#home) - Общи обекти в контекст. 80 класа, рамки и маски за сегментация
|
||||
|
||||
|
||||
|
||||
|
||||
## Метрики за разпознаване на обекти
|
||||
|
||||
|
|
@ -50,7 +50,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Докато за класификация на изображения е лесно да се измери колко добре се представя алгоритъмът, за разпознаване на обекти трябва да измерим както коректността на класа, така и прецизността на местоположението на предвидената рамка. За последното използваме така наречената **Пресечна площ спрямо обединение** (IoU), която измерва колко добре се припокриват две рамки (или две произволни области).
|
||||
|
||||
|
||||
|
||||
|
||||
> *Фигура 2 от [този отличен блог пост за IoU](https://pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/)*
|
||||
|
||||
|
|
@ -98,11 +98,11 @@ $$
|
|||
|
||||
[R-CNN](http://islab.ulsan.ac.kr/files/announcement/513/rcnn_pami.pdf) използва [Selective Search](http://www.huppelen.nl/publications/selectiveSearchDraft.pdf), за да генерира йерархична структура от региони ROI, които след това се обработват чрез CNN екстрактори на характеристики и SVM-класификатори, за да се определи класът на обекта, и линейна регресия за определяне на координатите на *рамката*. [Официална статия](https://arxiv.org/pdf/1506.01497v1.pdf)
|
||||
|
||||
|
||||
|
||||
|
||||
> *Изображение от van de Sande et al. ICCV’11*
|
||||
|
||||
|
||||
|
||||
|
||||
> *Изображения от [този блог](https://towardsdatascience.com/r-cnn-fast-r-cnn-faster-r-cnn-yolo-object-detection-algorithms-36d53571365e)
|
||||
|
||||
|
|
@ -110,7 +110,7 @@ $$
|
|||
|
||||
Този подход е подобен на R-CNN, но регионите се определят след прилагане на слоевете за конволюция.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [официалната статия](https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Girshick_Fast_R-CNN_ICCV_2015_paper.pdf), [arXiv](https://arxiv.org/pdf/1504.08083.pdf), 2015
|
||||
|
||||
|
|
@ -118,7 +118,7 @@ $$
|
|||
|
||||
Основната идея на този подход е да се използва невронна мрежа за предсказване на ROI - така наречената *Мрежа за предложения на региони*. [Статия](https://arxiv.org/pdf/1506.01497.pdf), 2016
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [официалната статия](https://arxiv.org/pdf/1506.01497.pdf)
|
||||
|
||||
|
|
@ -130,7 +130,7 @@ $$
|
|||
2. Характеристиките се обработват от **Position-Sensitive Score Map**. Всеки обект от $C$ класове се разделя на $k\times k$ региони, и обучаваме мрежата да предсказва части от обекти.
|
||||
3. За всяка част от $k\times k$ региони всички мрежи гласуват за класовете на обектите, и класът с максимален брой гласове се избира.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [официалната статия](https://arxiv.org/abs/1605.06409)
|
||||
|
||||
|
|
@ -141,7 +141,7 @@ YOLO е алгоритъм за разпознаване в реално вре
|
|||
* Изображението се разделя на $S\times S$ региони.
|
||||
* За всеки регион **CNN** предсказва $n$ възможни обекти, координати на *рамката* и *увереност*=*вероятност* * IoU.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [официалната статия](https://arxiv.org/abs/1506.02640)
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Компютърно зрение
|
||||
|
||||
|
||||
|
||||
|
||||
В тази секция ще научим за:
|
||||
|
||||
|
|
|
|||
|
|
@ -199,7 +199,7 @@
|
|||
"\n",
|
||||
"**Чанта с думи** (BoW) е най-често използваното традиционно векторно представяне. Всяка дума е свързана с индекс във вектора, а елементът на вектора съдържа броя на срещанията на дадена дума в конкретен документ.\n",
|
||||
"\n",
|
||||
" \n",
|
||||
" \n",
|
||||
"\n",
|
||||
"> **Note**: Можете също да мислите за BoW като за сума от всички едноразрядно кодирани вектори за отделните думи в текста.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -188,7 +188,7 @@
|
|||
"\n",
|
||||
"**Чанта с думи** (BoW) е най-простото за разбиране традиционно представяне на вектор. Всяка дума е свързана с индекс във вектора, а елементът на вектора съдържа броя на срещанията на всяка дума в даден документ.\n",
|
||||
"\n",
|
||||
" \n",
|
||||
" \n",
|
||||
"\n",
|
||||
"> **Note**: Можете също да мислите за BoW като сума от всички едно-горещо-кодирани вектори за отделните думи в текста.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -62,7 +62,7 @@
|
|||
"\n",
|
||||
"Използвайки слой за вграждане като първи слой в нашата мрежа, можем да преминем от модел на чанта с думи към модел на **чанта с вграждания**, където първо преобразуваме всяка дума в текста в съответното вграждане, а след това изчисляваме някаква агрегатна функция върху всички тези вграждания, като например `sum`, `average` или `max`.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Нашата невронна мрежа за класификация ще започне със слой за вграждане, след това слой за агрегиране и линеен класификатор върху него:\n"
|
||||
]
|
||||
|
|
@ -176,7 +176,7 @@
|
|||
"\n",
|
||||
"В предишната архитектура трябваше да запълним всички последователности до еднаква дължина, за да ги включим в минипартида. Това не е най-ефективният начин за представяне на последователности с променлива дължина - друг подход би бил използването на **вектор на отместванията**, който съдържа отместванията на всички последователности, съхранени в един голям вектор.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> **Note**: На изображението по-горе е показана последователност от символи, но в нашия пример работим с последователности от думи. Въпреки това, основният принцип на представяне на последователности с вектор на отмествания остава същият.\n",
|
||||
"\n",
|
||||
|
|
@ -311,7 +311,7 @@
|
|||
"\n",
|
||||
"CBoW е по-бърз, докато скип-грам е по-бавен, но се справя по-добре с представянето на редки думи.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"За да експериментираме с вграждания word2vec, предварително обучени върху набора от данни Google News, можем да използваме библиотеката **gensim**. По-долу намираме думите, които са най-близки до 'neural'.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -37,7 +37,7 @@
|
|||
"\n",
|
||||
"Като използваме embedding слой като първи слой в нашата мрежа, можем да преминем от модел bag-of-words към модел **embedding bag**, където първо преобразуваме всяка дума в текста в съответния embedding, а след това изчисляваме някаква агрегираща функция върху всички тези embeddings, като например `sum`, `average` или `max`.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Нашата невронна мрежа за класификация се състои от следните слоеве:\n",
|
||||
"\n",
|
||||
|
|
@ -283,7 +283,7 @@
|
|||
"\n",
|
||||
"CBoW е по-бърз, докато скип-грам е по-бавен, но се справя по-добре с представянето на редки думи.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"За да експериментираме с вграждането Word2Vec, предварително обучено върху набора от данни Google News, можем да използваме библиотеката **gensim**. По-долу намираме думите, които са най-близки до 'neural'.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Използвайки слой за вграждане като първи слой в нашата мрежа за класификация, можем да преминем от модел на чанта от думи към модел на **чанта от вграждания**, където първо преобразуваме всяка дума в текста в съответното вграждане и след това изчисляваме някаква агрегираща функция върху всички тези вграждания, като `sum`, `average` или `max`.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от автора
|
||||
|
||||
|
|
@ -40,7 +40,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
CBoW е по-бърз, докато скип-грам е по-бавен, но се справя по-добре с представянето на редки думи.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [тази статия](https://arxiv.org/pdf/1301.3781.pdf)
|
||||
|
||||
|
|
|
|||
|
|
@ -23,7 +23,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
* **Непрекъсната торба с думи** (CBoW), когато предсказваме средния токен $W_0$ в последователност от токени $W_{-N}$, ..., $W_N$.
|
||||
* **Skip-gram**, където предсказваме набор от съседни токени {$W_{-N},\dots, W_{-1}, W_1,\dots, W_N$} от средния токен $W_0$.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [тази статия](https://arxiv.org/pdf/1301.3781.pdf)
|
||||
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
За да уловим значението на текстовата последователност, трябва да използваме друга архитектура на невронна мрежа, наречена **рекурентна невронна мрежа** или RNN. В RNN подаваме изречението през мрежата символ по символ, а мрежата произвежда някакво **състояние**, което след това подаваме отново на мрежата заедно със следващия символ.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от автора
|
||||
|
||||
|
|
@ -61,7 +61,7 @@ LSTM мрежата е организирана по начин, подобен
|
|||
|
||||
Рекурентната мрежа, независимо дали е еднопосочна или двунаправлена, улавя определени модели в рамките на последователността и може да ги съхранява в състоянието или да ги предава като изход. Както при конволюционните мрежи, можем да изградим друг рекурентен слой върху първия, за да уловим модели на по-високо ниво и да изградим от модели на ниско ниво, извлечени от първия слой. Това ни води до понятието за **многослойна RNN**, която се състои от две или повече рекурентни мрежи, където изходът на предишния слой се подава на следващия слой като вход.
|
||||
|
||||
|
||||
|
||||
|
||||
*Изображение от [този чудесен пост](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) от Фернандо Лопес*
|
||||
|
||||
|
|
|
|||
|
|
@ -422,7 +422,7 @@
|
|||
"\n",
|
||||
"Рекурентната мрежа, независимо дали е еднопосочна или двунаправлена, улавя определени модели в рамките на последователността и може да ги съхранява в скрития вектор или да ги предава към изхода. Както при конволюционните мрежи, можем да изградим друг рекурентен слой върху първия, за да уловим модели на по-високо ниво, изградени от модели на ниско ниво, извлечени от първия слой. Това ни води до понятието **многослойна RNN**, която се състои от две или повече рекурентни мрежи, където изходът на предишния слой се подава като вход към следващия слой.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Снимка от [този чудесен пост](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) от Фернандо Лопес*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
"\n",
|
||||
"За да уловим значението на текстова последователност, ще използваме архитектура на невронна мрежа, наречена **рекурентна невронна мрежа** или RNN. Когато използваме RNN, подаваме изречението си през мрежата, една по една дума (токен), и мрежата произвежда някакво **състояние**, което след това подаваме отново на мрежата заедно със следващия токен.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Дадена входна последователност от токени $X_0,\\dots,X_n$, RNN създава последователност от блокове на невронната мрежа и обучава тази последователност от край до край, използвайки обратно разпространение. Всеки блок на мрежата приема двойка $(X_i,S_i)$ като вход и произвежда $S_{i+1}$ като резултат. Финалното състояние $S_n$ или изходът $Y_n$ се подава към линеен класификатор, за да се произведе резултатът. Всички блокове на мрежата споделят едни и същи тегла и се обучават от край до край с един цикъл на обратно разпространение.\n",
|
||||
"\n",
|
||||
|
|
@ -369,7 +369,7 @@
|
|||
"\n",
|
||||
"Рекурентните мрежи, еднопосочни или двунасочни, улавят модели в рамките на последователността и ги съхраняват в състояния или ги връщат като изход. Както при конволюционните мрежи, можем да изградим друг рекурентен слой след първия, за да уловим модели на по-високо ниво, изградени от модели на по-ниско ниво, извлечени от първия слой. Това ни води до понятието за **многослойна RNN**, която се състои от две или повече рекурентни мрежи, където изходът на предишния слой се предава на следващия слой като вход.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Снимка от [този чудесен пост](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) от Фернандо Лопес.*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -119,7 +119,7 @@
|
|||
"\n",
|
||||
"Начинът, по който ще обучим RNN да генерира текст, е следният. На всяка стъпка ще вземем последователност от символи с дължина `nchars` и ще помолим мрежата да генерира следващия изходен символ за всеки входен символ:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"В зависимост от конкретния сценарий, може да искаме да включим и някои специални символи, като например *край на последователността* `<eos>`. В нашия случай просто искаме да обучим мрежата за безкрайно генериране на текст, затова ще фиксираме размера на всяка последователност да бъде равен на `nchars` токени. Следователно, всеки тренировъчен пример ще се състои от `nchars` входове и `nchars` изходи (които са входната последователност, изместена с един символ наляво). Минипартидата ще се състои от няколко такива последователности.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -113,7 +113,7 @@
|
|||
"\n",
|
||||
"Начинът, по който ще обучим RNN да генерира новинарски заглавия, е следният. На всяка стъпка ще вземем едно заглавие, което ще бъде подадено в RNN, и за всеки входен символ ще поискаме от мрежата да генерира следващия изходен символ:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"За последния символ от нашата последователност ще поискаме от мрежата да генерира токен `<eos>`.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Това позволява различни невронни архитектури, които са показани на изображението по-долу:
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от блог пост [Unreasonable Effectiveness of Recurrent Neural Networks](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) от [Андрей Карпати](http://karpathy.github.io/)
|
||||
|
||||
|
|
@ -32,7 +32,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Ще обучим тази RNN да генерира текст стъпка по стъпка. На всяка стъпка ще вземем последователност от символи с дължина `nchars` и ще помолим мрежата да генерира следващия изходен символ за всеки входен символ:
|
||||
|
||||
|
||||
|
||||
|
||||
При генериране на текст (по време на инференция) започваме с някакъв **подсказка**, която се предава през RNN клетките, за да генерира междинното си състояние, и след това от това състояние започва генерирането. Генерираме един символ наведнъж и предаваме състоянието и генерирания символ на друга RNN клетка, за да генерира следващия, докато генерираме достатъчно символи.
|
||||
|
||||
|
|
|
|||
|
|
@ -20,13 +20,13 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
**Механизмите за внимание** предоставят начин за претегляне на контекстуалното влияние на всеки входен вектор върху всяка прогноза на изхода на RNN. Това се реализира чрез създаване на преки връзки между междинните състояния на входния RNN и изходния RNN. По този начин, когато генерираме изходния символ y<sub>t</sub>, ще вземем предвид всички скрити състояния на входа h<sub>i</sub>, с различни теглови коефициенти α<sub>t,i</sub>.
|
||||
|
||||
|
||||
|
||||
|
||||
> Моделът кодировач-декодировач с механизъм за добавено внимание в [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf), цитиран от [този блог пост](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html)
|
||||
|
||||
Матрицата за внимание {α<sub>i,j</sub>} представлява степента, до която определени входни думи играят роля в генерирането на дадена дума в изходната последователност. По-долу е пример за такава матрица:
|
||||
|
||||
|
||||
|
||||
|
||||
> Фигура от [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) (Фиг.3)
|
||||
|
||||
|
|
@ -66,7 +66,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Следващата стъпка е да уловим някои модели в рамките на нашата последователност. За да направят това, трансформерите използват механизъм за **самовнимание**, който по същество е внимание, приложено към една и съща последователност като вход и изход. Прилагането на самовнимание ни позволява да вземем предвид **контекста** в рамките на изречението и да видим кои думи са взаимосвързани. Например, това ни позволява да видим кои думи се отнасят до кореференции, като *it*, и също така да вземем контекста предвид:
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение от [Блогът на Google](https://research.googleblog.com/2017/08/transformer-novel-neural-network.html)
|
||||
|
||||
|
|
@ -91,7 +91,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
**BERT** (Bidirectional Encoder Representations from Transformers) е много голяма многослойна трансформерна мрежа с 12 слоя за *BERT-base* и 24 за *BERT-large*. Моделът първо се предварително обучава върху голям корпус от текстови данни (WikiPedia + книги) чрез неуправляемо обучение (предсказване на маскирани думи в изречение). По време на предварителното обучение моделът усвоява значителни нива на разбиране на езика, които след това могат да бъдат използвани с други набори от данни чрез фина настройка. Този процес се нарича **трансферно обучение**.
|
||||
|
||||
|
||||
|
||||
|
||||
> Изображение [източник](http://jalammar.github.io/illustrated-bert/)
|
||||
|
||||
|
|
|
|||
|
|
@ -12,12 +12,12 @@
|
|||
"\n",
|
||||
"**Механизмите на вниманието** предоставят начин за претегляне на контекстуалното влияние на всеки входен вектор върху всяка изходна прогноза на RNN. Това се реализира чрез създаване на преки връзки между междинните състояния на входната RNN и изходната RNN. По този начин, при генериране на изходен символ $y_t$, ще вземем предвид всички входни скрити състояния $h_i$, с различни теглови коефициенти $\\alpha_{t,i}$.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Моделът encoder-decoder с механизъм за добавено внимание в [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf), цитиран от [този блог пост](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html)*\n",
|
||||
"\n",
|
||||
"Матрицата на вниманието $\\{\\alpha_{i,j}\\}$ представлява степента, до която определени входни думи участват в генерирането на дадена дума в изходната последователност. По-долу е даден пример за такава матрица:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Фигура, взета от [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) (Фиг.3)*\n",
|
||||
"\n",
|
||||
|
|
@ -35,7 +35,7 @@
|
|||
"\n",
|
||||
"**BERT** (Bidirectional Encoder Representations from Transformers) е много голяма многослойна трансформерна мрежа с 12 слоя за *BERT-base* и 24 за *BERT-large*. Моделът първо се предварително обучава върху голям корпус от текстови данни (Wikipedia + книги) с помощта на неконтролирано обучение (предсказване на маскирани думи в изречение). По време на предварителното обучение моделът усвоява значително ниво на езиково разбиране, което след това може да бъде използвано с други набори от данни чрез фина настройка. Този процес се нарича **трансферно обучение**.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Съществуват много вариации на трансформерните архитектури, включително BERT, DistilBERT, BigBird, OpenGPT3 и други, които могат да бъдат фино настроени. Пакетът [HuggingFace](https://github.com/huggingface/) предоставя хранилище за обучение на много от тези архитектури с PyTorch.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -12,12 +12,12 @@
|
|||
"\n",
|
||||
"**Механизмите на внимание** предоставят начин за претегляне на контекстуалното влияние на всеки входен вектор върху всяка изходна прогноза на RNN. Това се реализира чрез създаване на преки връзки между междинните състояния на входната RNN и изходната RNN. По този начин, при генериране на изходен символ $y_t$, ще вземем предвид всички входни скрити състояния $h_i$, с различни теглови коефициенти $\\alpha_{t,i}$.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Моделът енкодер-декодер с механизъм за добавено внимание в [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf), цитиран от [този блог пост](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html)*\n",
|
||||
"\n",
|
||||
"Матрицата на внимание $\\{\\alpha_{i,j}\\}$ представлява степента, до която определени входни думи играят роля в генерирането на дадена дума в изходната последователност. По-долу е даден пример за такава матрица:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Фигура, взета от [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) (Фиг.3)*\n",
|
||||
"\n",
|
||||
|
|
@ -225,7 +225,7 @@
|
|||
"\n",
|
||||
"**BERT** (Bidirectional Encoder Representations from Transformers) е много голяма многослойна трансформаторна мрежа с 12 слоя за *BERT-base* и 24 за *BERT-large*. Моделът първо се предварително обучава върху голям корпус от текстови данни (WikiPedia + книги) чрез обучение без надзор (предсказване на маскирани думи в изречение). По време на предварителното обучение моделът усвоява значително ниво на езиково разбиране, което след това може да бъде използвано с други набори от данни чрез фина настройка. Този процес се нарича **трансферно обучение**.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Съществуват много вариации на трансформаторни архитектури, включително BERT, DistilBERT, BigBird, OpenGPT3 и други, които могат да бъдат фино настроени.\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ infant | O
|
|||
|
||||
Тъй като трябва да изградим едно към едно съответствие между токените и класовете, можем да обучим **много към много** невронен мрежов модел от тази картина:
|
||||
|
||||
|
||||
|
||||
|
||||
> *Изображение от [този блог пост](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) от [Андрей Карпати](http://karpathy.github.io/). Моделите за класификация на токени за NER съответстват на най-дясната архитектура на мрежата на тази картина.*
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Обработка на естествен език
|
||||
|
||||
|
||||
|
||||
|
||||
В тази секция ще се фокусираме върху използването на невронни мрежи за решаване на задачи, свързани с **обработката на естествен език (NLP)**. Съществуват много NLP проблеми, които искаме компютрите да могат да решават:
|
||||
|
||||
|
|
|
|||
|
|
@ -70,7 +70,7 @@ ask turtles [
|
|||
|
||||
След като отворите модела, ще бъдете отведени до основния екран на NetLogo. Ето примерен модел, който описва популацията на вълци и овце, при наличието на ограничени ресурси (трева).
|
||||
|
||||
|
||||
|
||||
|
||||
> Екранна снимка от Дмитрий Сошников
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Преглед
|
||||
|
||||
|
||||
|
||||
|
||||
> Рисунка от [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Основната идея на CLIP е да може да сравнява текстови подсказки с изображение и да определя доколко изображението съответства на подсказката.
|
||||
|
||||
|
||||
|
||||
|
||||
> *Снимка от [тази публикация в блог](https://openai.com/blog/clip/)*
|
||||
|
||||
|
|
@ -31,7 +31,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Да предположим, че трябва да класифицираме изображения между, например, котки, кучета и хора. В този случай можем да подадем на модела изображение и серия от текстови подсказки: "*снимка на котка*", "*снимка на куче*", "*снимка на човек*". В получения вектор от 3 вероятности просто трябва да изберем индекса с най-висока стойност.
|
||||
|
||||
|
||||
|
||||
|
||||
> *Снимка от [тази публикация в блог](https://openai.com/blog/clip/)*
|
||||
|
||||
|
|
@ -55,13 +55,13 @@ CLIP може също да се използва за **генериране н
|
|||
|
||||
Една от важните разлики между VQGAN и традиционните GAN е, че последните могат да произведат прилично изображение от всеки входен вектор, докато VQGAN е по-вероятно да произведе изображение, което не е кохерентно. Затова трябва допълнително да насочим процеса на създаване на изображението, което може да се направи с помощта на CLIP.
|
||||
|
||||
|
||||
|
||||
|
||||
За да генерираме изображение, съответстващо на текстова подсказка, започваме с някакъв случаен вектор за кодиране, който се подава през VQGAN, за да се произведе изображение. След това CLIP се използва за създаване на функция на загуба, която показва доколко изображението съответства на текстовата подсказка. Целта е да минимизираме тази загуба, използвайки обратна пропагация за настройка на параметрите на входния вектор.
|
||||
|
||||
Отлична библиотека, която реализира VQGAN+CLIP, е [Pixray](http://github.com/pixray/pixray).
|
||||
|
||||
 |  | 
|
||||
 |  | 
|
||||
----|----|----
|
||||
Снимка, генерирана от подсказка *близък акварелен портрет на млад мъж учител по литература с книга* | Снимка, генерирана от подсказка *близък маслен портрет на млада жена учител по компютърни науки с компютър* | Снимка, генерирана от подсказка *близък маслен портрет на възрастен мъж учител по математика пред черна дъска*
|
||||
|
||||
|
|
|
|||
|
|
@ -1,133 +1,133 @@
|
|||
<!--
|
||||
CO_OP_TRANSLATOR_METADATA:
|
||||
{
|
||||
"original_hash": "4ed9993bca581850c983c95d5a3f57eb",
|
||||
"translation_date": "2025-12-24T23:27:52+00:00",
|
||||
"original_hash": "14816e97d79b296c87811724f7785923",
|
||||
"translation_date": "2026-01-01T10:58:36+00:00",
|
||||
"source_file": "README.md",
|
||||
"language_code": "bn"
|
||||
}
|
||||
-->
|
||||
[](https://github.com/microsoft/AI-For-Beginners/blob/main/LICENSE)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/graphs/contributors/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/issues/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/pulls/)
|
||||
[](https://github.com/microsoft/AI-For-Beginners/blob/main/LICENSE)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/graphs/contributors/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/issues/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/pulls/)
|
||||
[](http://makeapullrequest.com)
|
||||
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/watchers/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/network/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/stargazers/)
|
||||
[](https://GitHub.com/microsoft/AI-For-Beginners/stargazers/)
|
||||
[](https://mybinder.org/v2/gh/microsoft/ai-for-beginners/HEAD)
|
||||
[](https://gitter.im/Microsoft/ai-for-beginners?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge)
|
||||
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
# শুরুদের জন্য কৃত্রিম বুদ্ধিমত্তা - একটি পাঠক্রম
|
||||
# শিক্ষানবিসদের জন্য কৃত্রিম বুদ্ধিমত্তা - একটি পাঠ্যক্রম
|
||||
|
||||
||
|
||||
||
|
||||
|:---:|
|
||||
| AI For Beginners - _স্কেচনোট লিখেছেন [@girlie_mac](https://twitter.com/girlie_mac)_ |
|
||||
| AI For Beginners - _স্কেচনোট by [@girlie_mac](https://twitter.com/girlie_mac)_ |
|
||||
|
||||
আমাদের 12-সপ্তাহ, 24-লেসনের পাঠক্রমের মাধ্যমে **কৃত্রিম বুদ্ধিমত্তা** (AI) এর জগৎ অন্বেষণ করুন! এতে ব্যবহারিক পাঠ, কুইজ এবং ল্যাব অন্তর্ভুক্ত রয়েছে। পাঠক্রমটি শুরুতেই উপযোগী এবং এতে TensorFlow ও PyTorch-এর মতো টুলস এবং AI-তে নৈতিকতা সম্পর্কিত বিষয়ক আলোচনা রয়েছে
|
||||
আমাদের 12-সপ্তাহ, 24-পাঠের পাঠ্যক্রমের মাধ্যমে **কৃত্রিম বুদ্ধিমত্তা** (AI) এর জগৎ অনুসন্ধান করুন! এতে ব্যবহারিক পাঠ, কুইজ এবং ল্যাব অন্তর্ভুক্ত রয়েছে। এই পাঠ্যক্রমটি শিক্ষানবিস-মৈত্রীপূর্ণ এবং এতে TensorFlow ও PyTorch-র মতো টুলগুলি এবং AI-এ নৈতিকতাসহ বিভিন্ন বিষয় কভার করা হয়েছে।
|
||||
|
||||
### 🌐 বহু-ভাষা সমর্থন
|
||||
### 🌐 বহুভাষী সমর্থন
|
||||
|
||||
#### GitHub Action-এর মাধ্যমে সমর্থিত (স্বয়ংক্রিয় & সর্বদা আপ-টু-ডেট)
|
||||
#### GitHub Action দ্বারা সমর্থিত (স্বয়ংক্রিয় ও সর্বদা আপ-টু-ডেট)
|
||||
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE START -->
|
||||
[আরবি](../ar/README.md) | [বাংলা](./README.md) | [বুলগেরিয়ান](../bg/README.md) | [বার্মিজ (মায়ানমার)](../my/README.md) | [চীনা (সরলীকৃত)](../zh/README.md) | [চীনা (পারম্পরিক, হংকং)](../hk/README.md) | [চীনা (পারম্পরিক, মাকাউ)](../mo/README.md) | [চীনা (পারম্পরিক, তাইওয়ান)](../tw/README.md) | [ক্রোয়েশিয়ান](../hr/README.md) | [চেক](../cs/README.md) | [ডেনিশ](../da/README.md) | [ডাচ](../nl/README.md) | [এস্তোনিয়ান](../et/README.md) | [ফিনিশ](../fi/README.md) | [ফরাসি](../fr/README.md) | [জার্মান](../de/README.md) | [গ্রিক](../el/README.md) | [হিব্রু](../he/README.md) | [হিন্দি](../hi/README.md) | [হাঙ্গেরিয়ান](../hu/README.md) | [ইন্দোনেশীয়](../id/README.md) | [ইতালিয়ান](../it/README.md) | [জাপানি](../ja/README.md) | [কন্নড়](../kn/README.md) | [কোরিয়ান](../ko/README.md) | [লিথুয়ানিয়ান](../lt/README.md) | [মালয়](../ms/README.md) | [মালায়ালম](../ml/README.md) | [মারাঠি](../mr/README.md) | [নেপালি](../ne/README.md) | [নাইজেরিয়ান পিজিন](../pcm/README.md) | [নরওয়েজিয়ান](../no/README.md) | [ফার্সি (পর্সি)](../fa/README.md) | [পোলিশ](../pl/README.md) | [পর্তুগিজ (ব্রাজিল)](../br/README.md) | [পর্তুগিজ (পর্তুগাল)](../pt/README.md) | [পাঞ্জাবি (গুরমুখি)](../pa/README.md) | [রোমানিয়ান](../ro/README.md) | [রাশিয়ান](../ru/README.md) | [সার্বিয়ান (সিরিলিক)](../sr/README.md) | [স্লোভাক](../sk/README.md) | [স্লোভেনিয়ান](../sl/README.md) | [স্প্যানিশ](../es/README.md) | [সোয়াহিলি](../sw/README.md) | [সুইডিশ](../sv/README.md) | [তাগালগ (ফিলিপিনো)](../tl/README.md) | [তামিল](../ta/README.md) | [তেলুগু](../te/README.md) | [থাই](../th/README.md) | [তুর্কি](../tr/README.md) | [ইউক্রেনীয়](../uk/README.md) | [উর্দু](../ur/README.md) | [ভিয়েতনামী](../vi/README.md)
|
||||
[আরবি](../ar/README.md) | [বাংলা](./README.md) | [বুলগেরীয়](../bg/README.md) | [বার্মিজ (মায়ানমার)](../my/README.md) | [চীনা (সরলীকৃত)](../zh/README.md) | [চীনা (প্রচলিত, হংকং)](../hk/README.md) | [চীনা (প্রচলিত, ম্যাকাও)](../mo/README.md) | [চীনা (প্রচলিত, তাইওয়ান)](../tw/README.md) | [ক্রোয়াটিয়ান](../hr/README.md) | [চেক](../cs/README.md) | [ড্যানিশ](../da/README.md) | [ডাচ](../nl/README.md) | [এস্তোনীয়](../et/README.md) | [ফিনিশ](../fi/README.md) | [ফরাসি](../fr/README.md) | [জার্মান](../de/README.md) | [গ্রিক](../el/README.md) | [হিব্রু](../he/README.md) | [হিন্দি](../hi/README.md) | [হাঙ্গেরিয়ান](../hu/README.md) | [ইন্দোনেশীয়](../id/README.md) | [ইতালীয়](../it/README.md) | [জাপানি](../ja/README.md) | [কন্নড়](../kn/README.md) | [কোরিয়ান](../ko/README.md) | [লিথুয়ানিয়ান](../lt/README.md) | [মালয়](../ms/README.md) | [মালায়ালম](../ml/README.md) | [মরাঠি](../mr/README.md) | [নেপালি](../ne/README.md) | [নাইজেরিয়ান পিজিন](../pcm/README.md) | [নরওয়েজিয়ান](../no/README.md) | [ফার্সি (ফার্সি)](../fa/README.md) | [পোলিশ](../pl/README.md) | [পর্তুগিজ (ব্রাজিল)](../br/README.md) | [পর্তুগিজ (পর্তুগাল)](../pt/README.md) | [পাঞ্জাবি (গুরুমুখী)](../pa/README.md) | [রোমানিয়ান](../ro/README.md) | [রাশিয়ান](../ru/README.md) | [সার্বিয়ান (সিরিলিক)](../sr/README.md) | [স্লোভাক](../sk/README.md) | [স্লোভেনীয়](../sl/README.md) | [স্প্যানিশ](../es/README.md) | [সুআহিলি](../sw/README.md) | [সুইডিশ](../sv/README.md) | [তাগালগ (ফিলিপিনো)](../tl/README.md) | [তামিল](../ta/README.md) | [তেলুগু](../te/README.md) | [থাই](../th/README.md) | [তুর্কি](../tr/README.md) | [ইউক্রেনীয়](../uk/README.md) | [উর্দু](../ur/README.md) | [ভিয়েতনামী](../vi/README.md)
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE END -->
|
||||
|
||||
**আপনি যদি অতিরিক্ত অনুবাদ ভাষা চান তা এখানে তালিকাভুক্ত করা আছে [here](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md)**
|
||||
**যদি আপনি অতিরিক্ত অনুবাদ চান তবে সমর্থিত ভাষাগুলির তালিকা [এখানে](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md) পাওয়া যাবে**
|
||||
|
||||
## কমিউনিটিতে যোগ দিন
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
## আপনি কী শিখবেন
|
||||
## আপনি যা শিখবেন
|
||||
|
||||
**[কোর্সের মাইন্ডম্যাপ](http://soshnikov.com/courses/ai-for-beginners/mindmap.html)**
|
||||
|
||||
এই পাঠক্রমে আপনি শিখবেন:
|
||||
এই পাঠ্যক্রমে, আপনি শিখবেন:
|
||||
|
||||
* কৃত্রিম বুদ্ধিমত্তার বিভিন্ন পদ্ধতি, যার মধ্যে "পুরনো" প্রতীকী পদ্ধতি যেমন **জ্ঞান উপস্থাপন** এবং তর্ককরণ (রিজনিং) ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence)) অন্তর্ভুক্ত।
|
||||
* **নিউরাল নেটওয়ার্ক** এবং **ডীপ লার্নিং**, যা আধুনিক AI-এর মূল। আমরা এই গুরুত্বপূর্ণ বিষয়গুলোর ধারণাগুলো দুইটি জনপ্রিয় ফ্রেমওয়ার্ক - [TensorFlow](http://Tensorflow.org) এবং [PyTorch](http://pytorch.org) - ব্যবহার করে কোড দিয়ে ব্যাখ্যা করব।
|
||||
* ছবি এবং টেক্সট নিয়ে কাজ করার জন্য **নিউরাল আর্কিটেকচার**। আমরা সাম্প্রতিক মডেলগুলো সম্পর্কে আলোচনা করব কিন্তু সম্ভবত সম্পূর্ণ সর্বাধুনিক (state-of-the-art) ব্যবস্থা পুরোপুরি কভার করা হবে না।
|
||||
* কম জনপ্রিয় AI পদ্ধতিসমূহ, যেমন **জেনেটিক অ্যালগরিদমস** এবং **মাল্টি-এজেন্ট সিস্টেম**।
|
||||
* কৃত্রিম বুদ্ধিমত্তার বিভিন্ন পন্থা, যার মধ্যে 'পুরনো' প্রতীকী পদ্ধতি — **জ্ঞান উপস্থাপন** এবং যুক্তি নির্ণয় ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence))।
|
||||
* **নিউরাল নেটওয়ার্ক** এবং **ডিপ লার্নিং**, যা আধুনিক AI-এর কেন্দ্রীয় অংশ। আমরা এই গুরুত্বপূর্ণ বিষয়গুলোর ধারণাগুলো দুইটি জনপ্রিয় ফ্রেমওয়ার্ক - [TensorFlow](http://Tensorflow.org) এবং [PyTorch](http://pytorch.org) - ব্যবহার করে কোডের মাধ্যমে ব্যাখ্যা করব।
|
||||
* চিত্র এবং টেক্সট নিয়ে কাজ করার জন্য **নিউরাল আর্কিটেকচার**। আমরা সাম্প্রতিক মডেলগুলি কভার করব, তবে হয়তো সর্বাধুনিক পর্যায় পর্যন্ত সবকিছু থাকবে না।
|
||||
* কম প্রচলিত AI পদ্ধতিসমূহ, যেমন **জেনেটিক অ্যালগরিদম** এবং **মাল্টি-এজেন্ট সিস্টেম**।
|
||||
|
||||
এই পাঠক্রমে আমরা কি কভার করব না:
|
||||
এই পাঠ্যক্রমে আমরা যা কভার করব না:
|
||||
|
||||
> [এই কোর্সের সকল অতিরিক্ত সম্পদ আমাদের Microsoft Learn সংগ্রহে খুঁজে পান](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)
|
||||
> [এই কোর্সের সমস্ত অতিরিক্ত সম্পদ আমাদের Microsoft Learn সংগ্রহে দেখুন](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)
|
||||
|
||||
* **বিজনেসে AI** ব্যবহারের ব্যবসায়িক কেস। Microsoft Learn-এ থাকা [Introduction to AI for business users](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পাথ বা [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum) নেওয়ার কথা বিবেচনা করুন, যা [INSEAD](https://www.insead.edu/) এর সহযোগিতায় তৈরি।
|
||||
* **ক্লাসিক মেশিন লার্নিং**, যা আমাদের [Machine Learning for Beginners Curriculum](http://github.com/Microsoft/ML-for-Beginners) এ ভালোভাবে বর্ণিত।
|
||||
* ব্যবহারিক AI অ্যাপ্লিকেশন যা **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)** ব্যবহার করে তৈরি। এর জন্য, আমরা Microsoft Learn-এর [vision](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [natural language processing](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Generative AI with Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** ইত্যাদি মডিউল নিয়ে শুরু করতে সুপারিশ করি।
|
||||
* নির্দিষ্ট ML **ক্লাউড ফ্রেমওয়ার্ক**, যেমন [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum), বা [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum)। [Build and operate machine learning solutions with Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) এবং [Build and Operate Machine Learning Solutions with Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পাথগুলো বিবেচনা করুন।
|
||||
* **কনভারসেশনাল AI** এবং **চ্যাট বটস**। এই বিষয়ে আলাদাভাবে [Create conversational AI solutions](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পাথ আছে, এবং বিস্তারিত জ্ঞানের জন্য আপনি [এই ব্লগ পোস্টটি](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/)ও দেখতে পারেন।
|
||||
* ডীপ লার্নিং-এর পেছনের **গভীর গণিত**। এর জন্য আমরা Ian Goodfellow, Yoshua Bengio এবং Aaron Courville-এর লেখা [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) বইটি সুপারিশ করব, যা অনলাইনে [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/) এ পাওয়া যায়।
|
||||
* **ব্যবসায় AI** ব্যবহার সংক্রান্ত ব্যবসায়িক কেসসমূহ। Microsoft Learn-এ [Introduction to AI for business users](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পথ নেওয়ার কথা বিবেচনা করুন, অথবা [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum), যা [INSEAD](https://www.insead.edu/)-এর সহযোগিতায় উন্নয়ন করা হয়েছে।
|
||||
* **ক্লাসিক মেশিন লার্নিং**, যা আমাদের [Machine Learning for Beginners Curriculum](http://github.com/Microsoft/ML-for-Beginners)-এ ভালভাবে বর্ণিত আছে।
|
||||
* **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)** ব্যবহার করে তৈরি বাস্তবগত AI অ্যাপ্লিকেশন। এর জন্য, আমরা সুপারিশ করব যে আপনি Microsoft Learn-এর [vision](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [natural language processing](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Generative AI with Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** এবং অন্যান্য মডিউল দিয়ে শুরু করতে।
|
||||
* নির্দিষ্ট এমএল **ক্লাউড ফ্রেমওয়ার্ক**, যেমন [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum), বা [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum)। বিবেচনা করুন [Build and operate machine learning solutions with Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) এবং [Build and Operate Machine Learning Solutions with Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পথ ব্যবহার করা।
|
||||
* **কথোপকথনমূলক AI** এবং **চ্যাট বট**। একটি পৃথক [Create conversational AI solutions](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পথ আছে, এবং বিস্তারিত জানতে আপনি [এই ব্লগ পোস্টটি](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/)ও দেখতে পারেন।
|
||||
* ডিপ লার্নিংয়ের পেছনের **গভীর গণিত**। এর জন্য, আমরা পরামর্শ দেব Ian Goodfellow, Yoshua Bengio এবং Aaron Courville-র লেখা [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) বইটি, যা অনলাইতেও [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/) পাওয়া যায়।
|
||||
|
||||
_AI in the Cloud_ বিষয়গুলোর নরম পরিচয়ের জন্য আপনি [Get started with artificial intelligence on Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পাথটি নিতে পারেন।
|
||||
ক্লাউড-ভিত্তিক _AI_ বিষয়গুলোর জন্য একটি সূক্ষ্ম পরিচয়ের জন্য আপনি [Get started with artificial intelligence on Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum) লার্নিং পাথ নিতে পারেন।
|
||||
|
||||
# বিষয়বস্তু
|
||||
|
||||
| | লেসনের লিঙ্ক | PyTorch/Keras/TensorFlow | Lab |
|
||||
| | পাঠের লিংক | PyTorch/Keras/TensorFlow | ল্যাব |
|
||||
| :-: | :------------------------------------------------------------------------------------------------------------------------------------------: | :---------------------------------------------------------------------------------------------: | ------------------------------------------------------------------------------ |
|
||||
| 0 | [Course Setup](./lessons/0-course-setup/setup.md) | [Setup Your Development Environment](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| I | [**AI-এর পরিচিতি**](./lessons/1-Intro/README.md) | | |
|
||||
| 01 | [AI এর পরিচিতি এবং ইতিহাস](./lessons/1-Intro/README.md) | - | - |
|
||||
| II | **প্রতীকী AI (Symbolic AI)** |
|
||||
| 02 | [জ্ঞান উপস্থাপন এবং এক্সপার্ট সিস্টেম](./lessons/2-Symbolic/README.md) | [Expert Systems](./lessons/2-Symbolic/Animals.ipynb) / [Ontology](./lessons/2-Symbolic/FamilyOntology.ipynb) /[Concept Graph](./lessons/2-Symbolic/MSConceptGraph.ipynb) | |
|
||||
| III | [**নিউরাল নেটওয়ার্কসে পরিচিতি**](./lessons/3-NeuralNetworks/README.md) |||
|
||||
| 0 | [কোর্স সেটআপ](./lessons/0-course-setup/setup.md) | [আপনার ডেভেলপমেন্ট পরিবেশ সেটআপ করুন](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| I | [**কৃত্রিম বুদ্ধিমত্তার পরিচিতি**](./lessons/1-Intro/README.md) | | |
|
||||
| 01 | [কৃত্রিম বুদ্ধিমত্তার পরিচয় ও ইতিহাস](./lessons/1-Intro/README.md) | - | - |
|
||||
| II | **প্রতীকী AI** |
|
||||
| 02 | [জ্ঞান উপস্থাপন এবং বিশেষজ্ঞ সিস্টেম](./lessons/2-Symbolic/README.md) | [বিশেষজ্ঞ সিস্টেম](./lessons/2-Symbolic/Animals.ipynb) / [অন্টলজি](./lessons/2-Symbolic/FamilyOntology.ipynb) /[কনসেপ্ট গ্রাফ](./lessons/2-Symbolic/MSConceptGraph.ipynb) | |
|
||||
| III | [**নিউরাল নেটওয়ার্কের পরিচিতি**](./lessons/3-NeuralNetworks/README.md) |||
|
||||
| 03 | [পারসেপট্রন](./lessons/3-NeuralNetworks/03-Perceptron/README.md) | [নোটবুক](./lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb) | [ল্যাব](./lessons/3-NeuralNetworks/03-Perceptron/lab/README.md) |
|
||||
| 04 | [মাল্টি-লেয়ার্ড পারসেপট্রন এবং আমাদের নিজস্ব ফ্রেমওয়ার্ক তৈরি করা](./lessons/3-NeuralNetworks/04-OwnFramework/README.md) | [নোটবুক](./lessons/3-NeuralNetworks/04-OwnFramework/OwnFramework.ipynb) | [ল্যাব](./lessons/3-NeuralNetworks/04-OwnFramework/lab/README.md) |
|
||||
| 05 | [ফ্রেমওয়ার্কগুলোর পরিচিতি (PyTorch/TensorFlow) এবং ওভারফিটিং](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [ল্যাব](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**কম্পিউটার ভিশন**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [Microsoft Azure-এ কম্পিউটার ভিশন অন্বেষণ করুন](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 05 | [ফ্রেমওয়ার্ক পরিচিতি (PyTorch/TensorFlow) এবং ওভারফিটিং](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [ল্যাব](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**কম্পিউটার ভিশন**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [কম্পিউটার ভিশন অন্বেষণ করুন Microsoft Azure-এ](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 06 | [কম্পিউটার ভিশনের পরিচিতি। OpenCV](./lessons/4-ComputerVision/06-IntroCV/README.md) | [নোটবুক](./lessons/4-ComputerVision/06-IntroCV/OpenCV.ipynb) | [ল্যাব](./lessons/4-ComputerVision/06-IntroCV/lab/README.md) |
|
||||
| 07 | [কনভলিউশনাল নিউরাল নেটওয়ার্কস](./lessons/4-ComputerVision/07-ConvNets/README.md) & [CNN আর্কিটেকচার](./lessons/4-ComputerVision/07-ConvNets/CNN_Architectures.md) | [PyTorch](./lessons/4-ComputerVision/07-ConvNets/ConvNetsPyTorch.ipynb) /[TensorFlow](./lessons/4-ComputerVision/07-ConvNets/ConvNetsTF.ipynb) | [ল্যাব](./lessons/4-ComputerVision/07-ConvNets/lab/README.md) |
|
||||
| 08 | [প্রি-ট্রেইনড নেটওয়ার্ক এবং ট্রান্সফার লার্নিং](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [প্রশিক্ষণের কৌশল](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [ল্যাব](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 09 | [অটোএনকোডার এবং VAE](./lessons/4-ComputerVision/09-Autoencoders/README.md) | [PyTorch](./lessons/4-ComputerVision/09-Autoencoders/AutoEncodersPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/09-Autoencoders/AutoencodersTF.ipynb) | |
|
||||
| 10 | [জেনারেটিভ অ্যাডভারসারিয়াল নেটওয়ার্ক এবং আর্টিস্টিক স্টাইল ট্রান্সফার](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 07 | [কনভলিউশনাল নিউরাল নেটওয়ার্ক](./lessons/4-ComputerVision/07-ConvNets/README.md) & [CNN আর্কিটেকচার্স](./lessons/4-ComputerVision/07-ConvNets/CNN_Architectures.md) | [PyTorch](./lessons/4-ComputerVision/07-ConvNets/ConvNetsPyTorch.ipynb) /[TensorFlow](./lessons/4-ComputerVision/07-ConvNets/ConvNetsTF.ipynb) | [ল্যাব](./lessons/4-ComputerVision/07-ConvNets/lab/README.md) |
|
||||
| 08 | [প্রি-ট্রেইন্ড নেটওয়ার্ক এবং ট্রান্সফার লার্নিং](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [Training Tricks](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [ল্যাব](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 09 | [অটোএনকোডার এবং VAEs](./lessons/4-ComputerVision/09-Autoencoders/README.md) | [PyTorch](./lessons/4-ComputerVision/09-Autoencoders/AutoEncodersPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/09-Autoencoders/AutoencodersTF.ipynb) | |
|
||||
| 10 | [জেনারেটিভ অ্যাডভারসারিয়াল নেটওয়ার্কস ও আর্টিস্টিক স্টাইল ট্রান্সফার](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 11 | [অবজেক্ট ডিটেকশন](./lessons/4-ComputerVision/11-ObjectDetection/README.md) | [TensorFlow](./lessons/4-ComputerVision/11-ObjectDetection/ObjectDetection.ipynb) | [ল্যাব](./lessons/4-ComputerVision/11-ObjectDetection/lab/README.md) |
|
||||
| 12 | [সেমান্টিক সেগমেন্টেশন। U-Net](./lessons/4-ComputerVision/12-Segmentation/README.md) | [PyTorch](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationPytorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationTF.ipynb) | |
|
||||
| 12 | [সেম্যানটিক সেগমেন্টেশন। U-Net](./lessons/4-ComputerVision/12-Segmentation/README.md) | [PyTorch](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationPytorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationTF.ipynb) | |
|
||||
| V | [**প্রাকৃতিক ভাষা প্রক্রিয়াকরণ**](./lessons/5-NLP/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-pytorch/?WT.mc_id=academic-77998-cacaste) /[TensorFlow](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-TensorFlow/?WT.mc_id=academic-77998-cacaste) | [Microsoft Azure-এ প্রাকৃতিক ভাষা প্রক্রিয়াকরণ অন্বেষণ করুন](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)|
|
||||
| 13 | [টেক্সট উপস্থাপন। BoW/TF-IDF](./lessons/5-NLP/13-TextRep/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationTF.ipynb) | |
|
||||
| 14 | [সেমান্টিক শব্দ এমবেডিং। Word2Vec এবং GloVe](./lessons/5-NLP/14-Embeddings/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsTF.ipynb) | |
|
||||
| 15 | [ভাষা মডেলিং। আপনার নিজের এমবেডিং প্রশিক্ষণ](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [ল্যাব](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| 16 | [রিকারেন্ট নিউরাল নেটওয়ার্ক (RNN)](./lessons/5-NLP/16-RNN/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNTF.ipynb) | |
|
||||
| 13 | [টেক্সট উপস্থাপনা। Bow/TF-IDF](./lessons/5-NLP/13-TextRep/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationTF.ipynb) | |
|
||||
| 14 | [সেম্যান্তিক ওয়ার্ড এমবেডিং। Word2Vec এবং GloVe](./lessons/5-NLP/14-Embeddings/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsTF.ipynb) | |
|
||||
| 15 | [ভাষা মডেলিং। নিজস্ব এমবেডিং প্রশিক্ষণ](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [ল্যাব](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| 16 | [রিকারেন্ট নিউরাল নেটওয়ার্ক](./lessons/5-NLP/16-RNN/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNTF.ipynb) | |
|
||||
| 17 | [জেনারেটিভ রিকারেন্ট নেটওয়ার্কস](./lessons/5-NLP/17-GenerativeNetworks/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativePyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativeTF.ipynb) | [ল্যাব](./lessons/5-NLP/17-GenerativeNetworks/lab/README.md) |
|
||||
| 18 | [ট্রান্সফর্মারস। BERT.](./lessons/5-NLP/18-Transformers/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersPyTorch.ipynb) /[TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersTF.ipynb) | |
|
||||
| 19 | [নামকৃত সত্তা সনাক্তকরণ](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [ল্যাব](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [বড় ভাষা মডেল, প্রম্পট প্রোগ্রামিং এবং কয়েক-শট টাস্ক](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| 18 | [ট্রান্সফর্মার্স। BERT.](./lessons/5-NLP/18-Transformers/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersPyTorch.ipynb) /[TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersTF.ipynb) | |
|
||||
| 19 | [নামকৃত সত্তা শনাক্তকরণ](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [ল্যাব](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [বৃহৎ ভাষা মডেল, প্রম্পট প্রোগ্রামিং এবং কয়েক-শট টাস্কস](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| VI | **অন্যান্য এআই কৌশল** || |
|
||||
| 21 | [জেনেটিক অ্যালগরিদম](./lessons/6-Other/21-GeneticAlgorithms/README.md) | [নোটবুক](./lessons/6-Other/21-GeneticAlgorithms/Genetic.ipynb) | |
|
||||
| 22 | [ডিপ রিইনফোর্সমেন্ট লার্নিং](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [ল্যাব](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 22 | [ডীপ রিইনফোর্সমেন্ট লার্নিং](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [ল্যাব](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 23 | [মাল্টি-এজেন্ট সিস্টেম](./lessons/6-Other/23-MultiagentSystems/README.md) | | |
|
||||
| VII | **এআই নৈতিকতা** | | |
|
||||
| 24 | [এআই নৈতিকতা এবং দায়িত্বশীল এআই](./lessons/7-Ethics/README.md) | [Microsoft Learn: দায়িত্বশীল AI নীতিমালা](https://docs.microsoft.com/learn/paths/responsible-ai-business-principles/?WT.mc_id=academic-77998-cacaste) | |
|
||||
| 24 | [এআই নৈতিকতা এবং দায়িত্বশীল এআই](./lessons/7-Ethics/README.md) | [Microsoft Learn: Responsible AI Principles](https://docs.microsoft.com/learn/paths/responsible-ai-business-principles/?WT.mc_id=academic-77998-cacaste) | |
|
||||
| IX | **অতিরিক্ত** | | |
|
||||
| 25 | [মাল্টি-মোডাল নেটওয়ার্ক, CLIP এবং VQGAN](./lessons/X-Extras/X1-MultiModal/README.md) | [নোটবুক](./lessons/X-Extras/X1-MultiModal/Clip.ipynb) | |
|
||||
| 25 | [মাল্টি-মোডাল নেটওয়ার্কস, CLIP এবং VQGAN](./lessons/X-Extras/X1-MultiModal/README.md) | [নোটবুক](./lessons/X-Extras/X1-MultiModal/Clip.ipynb) | |
|
||||
|
||||
## প্রতিটি পাঠে রয়েছে
|
||||
## প্রতি পাঠে রয়েছে
|
||||
|
||||
* পূর্ব-পঠন উপকরণ
|
||||
* চালনাযোগ্য Jupyter Notebooks, যেগুলো প্রায়ই নির্দিষ্ট ফ্রেমওয়ার্কের জন্য (**PyTorch** or **TensorFlow**). চালনাযোগ্য নোটবুকটিতে অনেক তাত্ত্বিক উপাদানও থাকে, তাই বিষয়টি বুঝতে আপনাকে নোটবুকের অন্তত একটি সংস্করণ (either PyTorch or TensorFlow) পড়ে যেতে হবে।
|
||||
* **Labs** নির্দিষ্ট বিষয়গুলিতে উপলব্ধ, যা আপনাকে শেখা বিষয়বস্তু একটি নির্দিষ্ট সমস্যায় প্রয়োগ করে দেখার সুযোগ দেয়।
|
||||
* কিছু অংশে সম্পর্কিত বিষয়গুলো কভার করা [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) মডিউলগুলোর লিঙ্ক রয়েছে।
|
||||
* পূর্ব-পাঠ্য উপকরণ
|
||||
* কার্যকর Jupyter নোটবুকসমূহ, যা প্রায়শই নির্দিষ্ট ফ্রেমওয়ার্ক (**PyTorch** বা **TensorFlow**) অনুযায়ী থাকে। কার্যকর নোটবুকটিতে প্রচুর তাত্ত্বিক উপকরণও থাকে, তাই বিষয়টি বুঝতে আপনাকে অন্তত একটি সংস্করণ (বা PyTorch বা TensorFlow) পড়ে দেখতে হবে।
|
||||
* **ল্যাবস** কিছু বিষয়ে উপলব্ধ, যা আপনাকে শেখা বিষয়টি একটি নির্দিষ্ট সমস্যায় প্রয়োগ করে দেখার সুযোগ দেয়।
|
||||
* কিছু সেকশনে সম্পর্কিত বিষয়াদি কভার করে এমন [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) মডিউলের লিংক রয়েছে।
|
||||
|
||||
## শুরু করা
|
||||
## Getting Started
|
||||
|
||||
### 🎯 AI-তে নতুন? এখান থেকে শুরু করুন!
|
||||
### 🎯 এআই-এ নতুন? এখান থেকেই শুরু করুন!
|
||||
|
||||
If you're completely new to AI and want quick, hands-on examples, check out our [**শুরু-উপযোগী উদাহরণ**](./examples/README.md)! These include:
|
||||
If you're completely new to AI and want quick, hands-on examples, check out our [**নবীন-বান্ধব উদাহরণ**](./examples/README.md)! These include:
|
||||
|
||||
- 🌟 **Hello AI World** - আপনার প্রথম AI প্রোগ্রাম (প্যাটার্ন শনাক্তকরণ)
|
||||
- 🧠 **Simple Neural Network** - শূন্য থেকে একটি নিউরাল নেটওয়ার্ক তৈরি করুন
|
||||
- 🖼️ **Image Classifier** - বিস্তারিত মন্তব্যসহ ইমেজগুলিকে শ্রেণীবদ্ধ করুন
|
||||
- 💬 **টেক্সট সেন্টিমেন্ট** - পজিটিভ/নেগেটিভ টেক্সট বিশ্লেষণ
|
||||
- 🌟 **Hello AI World** - Your first AI program (pattern recognition)
|
||||
- 🧠 **Simple Neural Network** - Build a neural network from scratch
|
||||
- 🖼️ **Image Classifier** - Classify images with detailed comments
|
||||
- 💬 **টেক্সট সেন্টিমেন্ট** - ইতিবাচক/নেতিবাচক টেক্সট বিশ্লেষণ
|
||||
|
||||
These examples are designed to help you understand AI concepts before diving into the full curriculum.
|
||||
|
||||
### 📚 পুরো পাঠ্যক্রম সেটআপ
|
||||
### 📚 সম্পূর্ণ পাঠ্যক্রম সেটআপ
|
||||
|
||||
- আমরা একটি [সেটআপ লেসন](./lessons/0-course-setup/setup.md) তৈরি করেছি যাতে আপনার ডেভেলপমেন্ট এনভায়রনমেন্ট সেটআপ করতে সহায়তা করা যায়। - শিক্ষকদের জন্য, আমরা আপনার জন্য একটি [পাঠ্যক্রম সেটআপ লেসন](./lessons/0-course-setup/for-teachers.md)ও তৈরি করেছি!
|
||||
- আমরা একটি [সেটআপ পাঠ](./lessons/0-course-setup/setup.md) তৈরি করেছি যাতে আপনার ডেভেলপমেন্ট পরিবেশ সেটআপ করতে সাহায্য করে। - শিক্ষকদের জন্যও, আমরা আপনার জন্য একটি [পাঠ্যক্রম সেটআপ পাঠ](./lessons/0-course-setup/for-teachers.md) তৈরি করেছি!
|
||||
- কিভাবে [VSCode বা Codepace-এ কোড চালাবেন](./lessons/0-course-setup/how-to-run.md)
|
||||
|
||||
Follow these steps:
|
||||
|
|
@ -138,15 +138,15 @@ Clone the Repository: `git clone https://github.com/microsoft/AI-For-Beginners.g
|
|||
|
||||
Don't forget to star (🌟) this repo to find it easier later.
|
||||
|
||||
## অন্যান্য শিক্ষার্থীদের সাথে পরিচিত হন
|
||||
## Meet other Learners
|
||||
|
||||
Join our [official AI Discord server](https://aka.ms/genai-discord?WT.mc_id=academic-105485-bethanycheum) to meet and network with other learners taking this course and get support.
|
||||
অন্য শিক্ষার্থীদের সাথে পরিচিত হওয়া এবং নেটওয়ার্ক গড়তে এবং সহায়তা পেতে আমাদের [অফিশিয়াল AI Discord সার্ভারে](https://aka.ms/genai-discord?WT.mc_id=academic-105485-bethanycheum) যোগ দিন।
|
||||
|
||||
If you have product feedback or questions whilst building visit our [Azure AI Foundry Developer Forum](https://aka.ms/foundry/forum)
|
||||
|
||||
## Quizzes
|
||||
|
||||
> **A note about quizzes**: All quizzes are contained in the Quiz-app folder in etc\quiz-app, or [Online Here](https://ff-quizzes.netlify.app/) They are linked from within the lessons the quiz app can be run locally or deployed to Azure; follow the instruction in the `quiz-app` folder. They are gradually being localized.
|
||||
> **কুইজ সম্পর্কে একটি নোট**: সকল কুইজগুলো etc\quiz-app-এ অবস্থিত Quiz-app ফোল্ডারে আছে, অথবা [অনলাইনে এখানে](https://ff-quizzes.netlify.app/)। এগুলো পাঠগুলোর মধ্যে লিঙ্ক করা আছে; কুইজ অ্যাপটি লোকালি চালানো যাবে অথবা Azure-এ ডেপ্লয় করা যাবে; `quiz-app` ফোল্ডারের নির্দেশনা অনুসরণ করুন। এগুলো ধীরে ধীরে লোকালাইজ করা হচ্ছে।
|
||||
|
||||
## Help Wanted
|
||||
|
||||
|
|
@ -156,9 +156,9 @@ Do you have suggestions or found spelling or code errors? Raise an issue or crea
|
|||
|
||||
* **✍️ প্রধান লেখক:** [Dmitry Soshnikov](http://soshnikov.com), PhD
|
||||
* **🔥 সম্পাদক:** [Jen Looper](https://twitter.com/jenlooper), PhD
|
||||
* **🎨 Sketchnote illustrator:** [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
* **✅ Quiz Creator:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 মূল অবদানকারী:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
* **🎨 স্কেচনোট চিত্রকর:** [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
* **✅ কুইজ নির্মাতা:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 মূল অবদানকারীরা:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
|
||||
## Other Curricula
|
||||
|
||||
|
|
@ -166,57 +166,57 @@ Our team produces other curricula! Check out:
|
|||
|
||||
<!-- CO-OP TRANSLATOR OTHER COURSES START -->
|
||||
### LangChain
|
||||
[](https://aka.ms/langchain4j-for-beginners)
|
||||
[](https://aka.ms/langchainjs-for-beginners?WT.mc_id=m365-94501-dwahlin)
|
||||
[](https://aka.ms/langchain4j-for-beginners)
|
||||
[](https://aka.ms/langchainjs-for-beginners?WT.mc_id=m365-94501-dwahlin)
|
||||
|
||||
---
|
||||
|
||||
### Azure / Edge / MCP / Agents
|
||||
[](https://github.com/microsoft/AZD-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/edgeai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mcp-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/ai-agents-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
### Azure / Edge / MCP / এজেন্টস
|
||||
[](https://github.com/microsoft/AZD-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/edgeai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mcp-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/ai-agents-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Generative AI Series
|
||||
[](https://github.com/microsoft/generative-ai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[-9333EA?style=for-the-badge&labelColor=E5E7EB&color=9333EA)](https://github.com/microsoft/Generative-AI-for-beginners-dotnet?WT.mc_id=academic-105485-koreyst)
|
||||
[-C084FC?style=for-the-badge&labelColor=E5E7EB&color=C084FC)](https://github.com/microsoft/generative-ai-for-beginners-java?WT.mc_id=academic-105485-koreyst)
|
||||
[-E879F9?style=for-the-badge&labelColor=E5E7EB&color=E879F9)](https://github.com/microsoft/generative-ai-with-javascript?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/generative-ai-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[-9333EA?style=for-the-badge&labelColor=E5E7EB&color=9333EA)](https://github.com/microsoft/Generative-AI-for-beginners-dotnet?WT.mc_id=academic-105485-koreyst)
|
||||
[-C084FC?style=for-the-badge&labelColor=E5E7EB&color=C084FC)](https://github.com/microsoft/generative-ai-for-beginners-java?WT.mc_id=academic-105485-koreyst)
|
||||
[-E879F9?style=for-the-badge&labelColor=E5E7EB&color=E879F9)](https://github.com/microsoft/generative-ai-with-javascript?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Core Learning
|
||||
[](https://aka.ms/ml-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/datascience-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ai-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/Security-101?WT.mc_id=academic-96948-sayoung)
|
||||
[](https://aka.ms/webdev-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/iot-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/xr-development-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ml-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/datascience-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ai-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/Security-101?WT.mc_id=academic-96948-sayoung)
|
||||
[](https://aka.ms/webdev-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/iot-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/xr-development-for-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
|
||||
---
|
||||
|
||||
### Copilot Series
|
||||
### কপাইলট সিরিজ
|
||||
[](https://aka.ms/GitHubCopilotAI?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mastering-github-copilot-for-dotnet-csharp-developers?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/CopilotAdventures?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/mastering-github-copilot-for-dotnet-csharp-developers?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://github.com/microsoft/CopilotAdventures?WT.mc_id=academic-105485-koreyst)
|
||||
<!-- CO-OP TRANSLATOR OTHER COURSES END -->
|
||||
|
||||
## সহায়তা পাওয়া
|
||||
## Getting Help
|
||||
|
||||
If you get stuck or have any questions about building AI apps. Join fellow learners and experienced developers in discussions about MCP. It's a supportive community where questions are welcome and knowledge is shared freely.
|
||||
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
If you have product feedback or errors while building visit:
|
||||
|
||||
[](https://aka.ms/foundry/forum)
|
||||
[](https://aka.ms/foundry/forum)
|
||||
|
||||
---
|
||||
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER START -->
|
||||
অস্বীকৃতি:
|
||||
এই নথিটি AI অনুবাদ সেবা Co‑op Translator (https://github.com/Azure/co-op-translator) ব্যবহার করে অনুবাদ করা হয়েছে। যদিও আমরা যথাসাধ্য সঠিকতা নিশ্চিত করার চেষ্টা করি, অনুগ্রহ করে জানুন যে স্বয়ংক্রিয় অনুবাদে ত্রুটি বা ভুল থাকতে পারে। মূল ভাষায় থাকা নথিটিকেই প্রামাণিক উৎস হিসেবে বিবেচনা করা উচিত। গুরুত্বপূর্ণ তথ্যের জন্য পেশাদার মানব অনুবাদ করা উচিৎ। এই অনুবাদ ব্যবহারের ফলে হওয়া কোনও ভুল বোঝাবুঝি বা ভুল ব্যাখ্যার জন্য আমরা দায়ী নই।
|
||||
এই নথিটি AI অনুবাদ সেবা Co-op Translator (https://github.com/Azure/co-op-translator) ব্যবহার করে অনুবাদ করা হয়েছে। যদিও আমরা যথাসাধ্য সঠিকতা বজায় রাখার চেষ্টা করি, অনুগ্রহ করে মনে রাখুন যে স্বয়ংক্রিয় অনুবাদে ত্রুটি বা অসঙ্গতি থাকতে পারে। মূল ভাষায় থাকা নথিটিকেই প্রাধান্যপূর্ণ এবং নির্ভরযোগ্য উৎস হিসেবে গণ্য করা উচিত। গুরুত্বর্পূৰ্ণ তথ্যের ক্ষেত্রে পেশাদার মানব অনুবাদের পরামর্শ দেওয়া হয়। এই অনুবাদ ব্যবহারের ফলে উদ্ভূত কোনো ভুল বোঝাবুঝি বা ভ্রান্ত ব্যাখ্যার জন্য আমরা দায়ী নই।
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER END -->
|
||||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# AI পরিচিতি
|
||||
|
||||
|
||||
|
||||
|
||||
> স্কেচনোট: [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
মূলত, কম্পিউটার আবিষ্কার করেছিলেন [Charles Babbage](https://en.wikipedia.org/wiki/Charles_Babbage) সংখ্যা নিয়ে কাজ করার জন্য, একটি সুস্পষ্ট প্রক্রিয়া অনুসরণ করে - একটি অ্যালগরিদম। আধুনিক কম্পিউটার, যদিও ১৯শ শতাব্দীতে প্রস্তাবিত মূল মডেলের তুলনায় অনেক বেশি উন্নত, তবুও নিয়ন্ত্রিত গণনার একই ধারণা অনুসরণ করে। তাই, যদি আমরা জানি লক্ষ্য অর্জনের জন্য প্রয়োজনীয় সঠিক ধাপগুলোর ক্রম, তাহলে কম্পিউটারকে কিছু করতে প্রোগ্রাম করা সম্ভব।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি: [Vickie Soshnikova](http://twitter.com/vickievalerie)
|
||||
|
||||
|
|
@ -46,7 +46,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
**[বুদ্ধিমত্তা](https://en.wikipedia.org/wiki/Intelligence)** শব্দটি নিয়ে কাজ করার সময় একটি সমস্যা হলো এই শব্দটির কোনো স্পষ্ট সংজ্ঞা নেই। কেউ যুক্তি করতে পারে যে বুদ্ধিমত্তা **অমূর্ত চিন্তা** বা **আত্ম-সচেতনতার** সাথে সংযুক্ত, কিন্তু আমরা এটি সঠিকভাবে সংজ্ঞায়িত করতে পারি না।
|
||||
|
||||
|
||||
|
||||
|
||||
> [ছবি](https://unsplash.com/photos/75715CVEJhI): [Amber Kipp](https://unsplash.com/@sadmax) Unsplash থেকে
|
||||
|
||||
|
|
@ -98,13 +98,13 @@ AGI নিয়ে কথা বলার সময় আমাদের এমন
|
|||
|
||||
> | ML সম্পর্কে কী? | |
|
||||
> |--------------|-----------|
|
||||
> | কৃত্রিম বুদ্ধিমত্তার অংশ যা কম্পিউটারকে কিছু ডেটার ভিত্তিতে একটি সমস্যা সমাধানে শেখানোর উপর ভিত্তি করে তাকে **মেশিন লার্নিং** বলা হয়। আমরা এই কোর্সে ক্লাসিক্যাল মেশিন লার্নিং বিবেচনা করব না - আমরা আপনাকে একটি পৃথক [Machine Learning for Beginners](http://aka.ms/ml-beginners) কারিকুলামে রেফার করছি। |  |
|
||||
> | কৃত্রিম বুদ্ধিমত্তার অংশ যা কম্পিউটারকে কিছু ডেটার ভিত্তিতে একটি সমস্যা সমাধানে শেখানোর উপর ভিত্তি করে তাকে **মেশিন লার্নিং** বলা হয়। আমরা এই কোর্সে ক্লাসিক্যাল মেশিন লার্নিং বিবেচনা করব না - আমরা আপনাকে একটি পৃথক [Machine Learning for Beginners](http://aka.ms/ml-beginners) কারিকুলামে রেফার করছি। |  |
|
||||
|
||||
## AI এর সংক্ষিপ্ত ইতিহাস
|
||||
|
||||
কৃত্রিম বুদ্ধিমত্তা একটি ক্ষেত্র হিসেবে শুরু হয়েছিল বিংশ শতাব্দীর মাঝামাঝি। প্রথমদিকে, প্রতীকী যুক্তি একটি প্রভাবশালী পদ্ধতি ছিল, এবং এটি কিছু গুরুত্বপূর্ণ সাফল্য অর্জন করেছিল, যেমন বিশেষজ্ঞ সিস্টেম – কম্পিউটার প্রোগ্রাম যা কিছু সীমিত সমস্যার ক্ষেত্রে একজন বিশেষজ্ঞের মতো কাজ করতে সক্ষম ছিল। তবে, শীঘ্রই এটি স্পষ্ট হয়ে যায় যে এই পদ্ধতি ভালোভাবে স্কেল করে না। একজন বিশেষজ্ঞ থেকে জ্ঞান বের করা, কম্পিউটারে উপস্থাপন করা, এবং সেই জ্ঞানভাণ্ডারকে সঠিক রাখা একটি অত্যন্ত জটিল কাজ এবং অনেক ক্ষেত্রে ব্যবহারিকভাবে খুব ব্যয়বহুল। এটি ১৯৭০-এর দশকে তথাকথিত [AI Winter](https://en.wikipedia.org/wiki/AI_winter) এর দিকে নিয়ে যায়।
|
||||
|
||||
<img alt="AI এর সংক্ষিপ্ত ইতিহাস" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a0264357672b0884cf3a220fbafe35c65d70b2c3805f7bf5e.bn.png" width="70%"/>
|
||||
<img alt="AI এর সংক্ষিপ্ত ইতিহাস" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a.bn.png" width="70%"/>
|
||||
|
||||
> ছবি: [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
|
|||
|
|
@ -12,7 +12,7 @@
|
|||
"\n",
|
||||
"এই উদাহরণে, আমরা একটি সহজ জ্ঞান-ভিত্তিক সিস্টেম বাস্তবায়ন করব যা কিছু শারীরিক বৈশিষ্ট্যের উপর ভিত্তি করে একটি প্রাণী নির্ধারণ করবে। সিস্টেমটি নিম্নলিখিত AND-OR গাছ দ্বারা উপস্থাপন করা যেতে পারে (এটি পুরো গাছের একটি অংশ, আমরা সহজেই আরও কিছু নিয়ম যোগ করতে পারি):\n",
|
||||
"\n",
|
||||
"\n"
|
||||
"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# জ্ঞান উপস্থাপন এবং বিশেষজ্ঞ সিস্টেম
|
||||
|
||||
|
||||
|
||||
|
||||
> স্কেচনোট [Tomomi Imura](https://twitter.com/girlie_mac) দ্বারা
|
||||
|
||||
|
|
@ -41,7 +41,7 @@ AI-এর প্রাথমিক দিনগুলোতে, বুদ্ধ
|
|||
|
||||
তাই, **জ্ঞান উপস্থাপনের** সমস্যাটি হল কম্পিউটারের ভিতরে ডেটার আকারে জ্ঞান উপস্থাপন করার কার্যকর উপায় খুঁজে বের করা, যাতে এটি স্বয়ংক্রিয়ভাবে ব্যবহারযোগ্য হয়। এটি একটি স্পেকট্রামের মতো দেখা যেতে পারে:
|
||||
|
||||
|
||||
|
||||
|
||||
> চিত্র [Dmitry Soshnikov](http://soshnikov.com) দ্বারা
|
||||
|
||||
|
|
@ -94,7 +94,7 @@ Block Syntax | Indent | | |
|
|||
|
||||
সিম্বলিক AI-এর প্রাথমিক সাফল্যগুলোর মধ্যে একটি ছিল তথাকথিত **বিশেষজ্ঞ সিস্টেম** - কম্পিউটার সিস্টেম যা একটি সীমিত সমস্যার ক্ষেত্রে একজন বিশেষজ্ঞের মতো কাজ করার জন্য ডিজাইন করা হয়েছিল। এগুলো একটি **জ্ঞানভিত্তি** এবং একটি **ইনফারেন্স ইঞ্জিন** নিয়ে গঠিত ছিল যা এর উপর যুক্তি করত।
|
||||
|
||||
 | 
|
||||
 | 
|
||||
---------------------------------------------|------------------------------------------------
|
||||
মানব স্নায়ুতন্ত্রের সরলীকৃত কাঠামো | জ্ঞানভিত্তিক সিস্টেমের স্থাপত্য
|
||||
|
||||
|
|
@ -106,7 +106,7 @@ Block Syntax | Indent | | |
|
|||
|
||||
একটি উদাহরণ হিসেবে, চলুন একটি বিশেষজ্ঞ সিস্টেমের কথা বিবেচনা করি যা শারীরিক বৈশিষ্ট্যের উপর ভিত্তি করে একটি প্রাণী নির্ধারণ করে:
|
||||
|
||||
|
||||
|
||||
|
||||
> চিত্র [Dmitry Soshnikov](http://soshnikov.com) দ্বারা
|
||||
|
||||
|
|
|
|||
|
|
@ -1259,7 +1259,7 @@
|
|||
"* প্রশিক্ষণ ক্ষতি কম - মডেল প্রশিক্ষণ ডেটাকে ভালোভাবে অনুমান করতে পারে, কারণ এর যথেষ্ট প্রকাশক্ষমতা রয়েছে।\n",
|
||||
"* যাচাইকরণ ক্ষতি প্রশিক্ষণ ক্ষতির তুলনায় অনেক বেশি হতে পারে এবং প্রশিক্ষণের সময় বৃদ্ধি পেতে পারে - কারণ মডেল প্রশিক্ষণ পয়েন্টগুলোকে \"মনে রাখে\" এবং \"সামগ্রিক চিত্র\" হারিয়ে ফেলে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> এই ছবিতে, `x` প্রশিক্ষণ ডেটাকে নির্দেশ করে, `o` - যাচাইকরণ ডেটা। বামদিকে - লিনিয়ার মডেল (এক-লেয়ার), এটি ডেটার প্রকৃতিকে বেশ ভালোভাবে অনুমান করে। ডানদিকে - ওভারফিটেড মডেল, মডেল প্রশিক্ষণ ডেটাকে পুরোপুরি ভালোভাবে অনুমান করে, কিন্তু অন্য কোনো ডেটার ক্ষেত্রে অর্থহীন হয়ে যায় (যাচাইকরণ ত্রুটি খুব বেশি)।\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ High-level API| [Keras](IntroKeras.ipynb) | *PyTorch Lightning*
|
|||
|
||||
নিচের ৫টি বিন্দু (গ্রাফে `x` দ্বারা চিহ্নিত) আনুমানিক করার সমস্যাটি বিবেচনা করুন:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-------------------------|--------------------------
|
||||
**লিনিয়ার মডেল, ২টি প্যারামিটার** | **নন-লিনিয়ার মডেল, ৭টি প্যারামিটার**
|
||||
প্রশিক্ষণ ত্রুটি = ৫.৩ | প্রশিক্ষণ ত্রুটি = ০
|
||||
|
|
@ -79,7 +79,7 @@ High-level API| [Keras](IntroKeras.ipynb) | *PyTorch Lightning*
|
|||
|
||||
উপরের গ্রাফ থেকে আপনি দেখতে পাচ্ছেন, ওভারফিটিং খুব কম প্রশিক্ষণ ত্রুটি এবং উচ্চ ভ্যালিডেশন ত্রুটি দ্বারা সনাক্ত করা যায়। সাধারণত প্রশিক্ষণের সময় আমরা দেখতে পাবো প্রশিক্ষণ এবং ভ্যালিডেশন ত্রুটি উভয়ই কমতে শুরু করে, এবং তারপর কোনো এক সময় ভ্যালিডেশন ত্রুটি কমা বন্ধ করে এবং বাড়তে শুরু করে। এটি ওভারফিটিংয়ের একটি চিহ্ন হবে এবং এটি নির্দেশ করবে যে আমাদের সম্ভবত এই সময়ে প্রশিক্ষণ বন্ধ করা উচিত (অথবা অন্তত মডেলের একটি স্ন্যাপশট নেওয়া উচিত)।
|
||||
|
||||
|
||||
|
||||
|
||||
## কিভাবে ওভারফিটিং প্রতিরোধ করবেন
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# নিউরাল নেটওয়ার্কের পরিচিতি
|
||||
|
||||
|
||||
|
||||
|
||||
যেমনটি আমরা পরিচিতিতে আলোচনা করেছি, বুদ্ধিমত্তা অর্জনের একটি উপায় হল একটি **কম্পিউটার মডেল** বা একটি **কৃত্রিম মস্তিষ্ক** প্রশিক্ষণ দেওয়া। ২০শ শতাব্দীর মাঝামাঝি থেকে গবেষকরা বিভিন্ন গাণিতিক মডেল চেষ্টা করেছেন, এবং সাম্প্রতিক বছরগুলোতে এই দিকটি অত্যন্ত সফল প্রমাণিত হয়েছে। মস্তিষ্কের এই ধরনের গাণিতিক মডেলগুলোকে **নিউরাল নেটওয়ার্ক** বলা হয়।
|
||||
|
||||
|
|
@ -36,13 +36,13 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
জীববিজ্ঞান থেকে আমরা জানি যে আমাদের মস্তিষ্ক নিউরাল কোষ (নিউরন) নিয়ে গঠিত, প্রতিটি কোষের একাধিক "ইনপুট" (ডেনড্রাইট) এবং একটি "আউটপুট" (অ্যাক্সন) থাকে। ডেনড্রাইট এবং অ্যাক্সন উভয়ই বৈদ্যুতিক সংকেত পরিচালনা করতে পারে, এবং তাদের মধ্যে সংযোগগুলো — যেগুলো সিন্যাপস নামে পরিচিত — বিভিন্ন মাত্রার পরিবাহিতা প্রদর্শন করতে পারে, যা নিউরোট্রান্সমিটার দ্বারা নিয়ন্ত্রিত হয়।
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
আসল নিউরন *([উইকিপিডিয়া থেকে ইমেজ](https://en.wikipedia.org/wiki/Synapse#/media/File:SynapseSchematic_lines.svg))* | কৃত্রিম নিউরন *(লেখকের তৈরি ইমেজ)*
|
||||
|
||||
তাই, একটি নিউরনের সবচেয়ে সহজ গাণিতিক মডেলটি কয়েকটি ইনপুট X<sub>1</sub>, ..., X<sub>N</sub> এবং একটি আউটপুট Y, এবং একটি সিরিজ ওজন W<sub>1</sub>, ..., W<sub>N</sub> নিয়ে গঠিত। আউটপুটটি নিম্নরূপ গণনা করা হয়:
|
||||
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd767313e067719f400cec4b0e5090239c3e997c29f6789d4c3c263.bn.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd76731.bn.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
|
||||
যেখানে f হল কিছু অরৈখিক **অ্যাক্টিভেশন ফাংশন**।
|
||||
|
||||
|
|
|
|||
|
|
@ -73,14 +73,14 @@ OpenCV ব্যবহার করে ভিডিও ফ্রেম-বাই
|
|||
|
||||
* **ব্রেইল বইয়ের একটি ছবির প্রি-প্রসেসিং**। আমরা দেখিয়েছি কীভাবে থ্রেশহোল্ডিং, ফিচার ডিটেকশন, পার্সপেক্টিভ ট্রান্সফরমেশন এবং NumPy ম্যানিপুলেশন ব্যবহার করে ব্রেইল প্রতীকগুলোকে আলাদা করা যায়, যা পরে নিউরাল নেটওয়ার্ক দ্বারা শ্রেণীবদ্ধ করা হবে।
|
||||
|
||||
 |  | 
|
||||
 |  | 
|
||||
----|-----|-----
|
||||
|
||||
> ছবি [OpenCV.ipynb](OpenCV.ipynb) থেকে
|
||||
|
||||
* **ভিডিওতে ফ্রেম ডিফারেন্স ব্যবহার করে গতিবিধি সনাক্তকরণ**। যদি ক্যামেরা স্থির থাকে, তাহলে ক্যামেরা ফিডের ফ্রেমগুলো একে অপরের সাথে বেশ মিল থাকবে। যেহেতু ফ্রেমগুলো অ্যারে হিসেবে উপস্থাপিত হয়, দুটি পরবর্তী ফ্রেমের জন্য সেই অ্যারেগুলোকে বিয়োগ করলেই আমরা পিক্সেল পার্থক্য পাব, যা স্থির ফ্রেমের জন্য কম হবে এবং ছবিতে উল্লেখযোগ্য গতিবিধি থাকলে বেশি হবে।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [OpenCV.ipynb](OpenCV.ipynb) থেকে
|
||||
|
||||
|
|
@ -89,7 +89,7 @@ OpenCV ব্যবহার করে ভিডিও ফ্রেম-বাই
|
|||
- **ডেন্স অপটিক্যাল ফ্লো** প্রতিটি পিক্সেলের জন্য একটি ভেক্টর ক্ষেত্র গণনা করে যা দেখায় এটি কোথায় স্থানান্তরিত হচ্ছে।
|
||||
- **স্পার্স অপটিক্যাল ফ্লো** ছবিতে কিছু বৈশিষ্ট্যপূর্ণ বৈশিষ্ট্য (যেমন: প্রান্ত) গ্রহণ করে এবং ফ্রেম থেকে ফ্রেমে তাদের গতিপথ তৈরি করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [OpenCV.ipynb](OpenCV.ipynb) থেকে
|
||||
|
||||
|
|
|
|||
|
|
@ -13,11 +13,11 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
VGG-16 একটি নেটওয়ার্ক যা ২০১৪ সালে ImageNet টপ-৫ শ্রেণীবিন্যাসে ৯২.৭% সঠিকতা অর্জন করেছিল। এর স্তর কাঠামো নিম্নরূপ:
|
||||
|
||||
|
||||
|
||||
|
||||
যেমনটি আপনি দেখতে পাচ্ছেন, VGG একটি ঐতিহ্যবাহী পিরামিড আর্কিটেকচার অনুসরণ করে, যা কনভোলিউশন-পুলিং স্তরের একটি ক্রম।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [Researchgate](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493) থেকে
|
||||
|
||||
|
|
|
|||
|
|
@ -260,7 +260,7 @@
|
|||
"\n",
|
||||
"সুতরাং, একটি সাধারণ CNN-এ কয়েকটি কনভল্যুশনাল লেয়ার থাকবে, যার মধ্যে পুলিং লেয়ার থাকবে যা ছবির মাত্রা কমিয়ে দেবে। আমরা ফিল্টারের সংখ্যাও বাড়াবো, কারণ প্যাটার্ন যত উন্নত হয় - তত বেশি সম্ভাব্য আকর্ষণীয় সংমিশ্রণ থাকে যা আমাদের খুঁজে বের করতে হবে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"স্থানিক মাত্রা কমানো এবং ফিচার/ফিল্টারের মাত্রা বাড়ানোর কারণে, এই আর্কিটেকচারকে **পিরামিড আর্কিটেকচার**ও বলা হয়।\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -358,7 +358,7 @@
|
|||
"\n",
|
||||
"এইভাবে, একটি সাধারণ CNN-এ কয়েকটি কনভোলিউশনাল লেয়ার থাকে, যার মধ্যে পুলিং লেয়ারগুলো ছবির মাত্রা কমানোর জন্য ব্যবহৃত হয়। আমরা ফিল্টারের সংখ্যা বাড়িয়ে দিই, কারণ প্যাটার্নগুলো আরও উন্নত হওয়ার সাথে সাথে আরও বেশি সম্ভাব্য আকর্ষণীয় সংমিশ্রণ থাকে যা আমাদের খুঁজে বের করতে হয়।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"স্থানিক মাত্রা কমানো এবং ফিচার/ফিল্টার মাত্রা বাড়ানোর কারণে, এই আর্কিটেকচারকে **পিরামিড আর্কিটেকচার** বলা হয়।\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -17,7 +17,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
প্যাটার্ন বের করার জন্য, আমরা **কনভোলিউশনাল ফিল্টার** ধারণাটি ব্যবহার করব। আপনি জানেন, একটি ছবি 2D-ম্যাট্রিক্স বা রঙের গভীরতা সহ 3D-টেনসর দ্বারা উপস্থাপিত হয়। একটি ফিল্টার প্রয়োগ করার অর্থ হলো আমরা একটি তুলনামূলকভাবে ছোট **ফিল্টার কের্নেল** ম্যাট্রিক্স নিই, এবং মূল ছবির প্রতিটি পিক্সেলের জন্য আমরা প্রতিবেশী পয়েন্টগুলির সাথে ওজনযুক্ত গড় গণনা করি। আমরা এটি একটি ছোট জানালা হিসেবে দেখতে পারি যা পুরো ছবির উপর দিয়ে স্লাইড করে এবং ফিল্টার কের্নেল ম্যাট্রিক্সের ওজন অনুযায়ী সমস্ত পিক্সেল গড় করে।
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
|
||||
> ছবি: দিমিত্রি সশনিকভ
|
||||
|
|
@ -38,7 +38,7 @@ CNN যেভাবে কাজ করে তা নিম্নলিখিত
|
|||
* আমরা নেটওয়ার্কটিকে এমনভাবে ডিজাইন করতে পারি যাতে ফিল্টারগুলি স্বয়ংক্রিয়ভাবে প্রশিক্ষিত হয়
|
||||
* আমরা একই পদ্ধতি ব্যবহার করে উচ্চ-স্তরের বৈশিষ্ট্যগুলিতে প্যাটার্ন খুঁজে বের করতে পারি, শুধুমাত্র মূল ছবিতে নয়। সুতরাং CNN বৈশিষ্ট্য বের করার কাজ বৈশিষ্ট্যের একটি শ্রেণিবিন্যাসে কাজ করে, যা নিম্ন-স্তরের পিক্সেল সংমিশ্রণ থেকে শুরু করে ছবির অংশগুলির উচ্চ-স্তরের সংমিশ্রণে পৌঁছায়।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি: [হিসলপ-লিঞ্চের গবেষণাপত্র](https://www.semanticscholar.org/paper/Computer-vision-based-pedestrian-trajectory-Hislop-Lynch/26e6f74853fc9bbb7487b06dc2cf095d36c9021d), তাদের [গবেষণার উপর ভিত্তি করে](https://dl.acm.org/doi/abs/10.1145/1553374.1553453)
|
||||
|
||||
|
|
@ -55,9 +55,9 @@ CNN যেভাবে কাজ করে তা নিম্নলিখিত
|
|||
|
||||
উদাহরণস্বরূপ, আসুন VGG-16-এর আর্কিটেকচার দেখি, একটি নেটওয়ার্ক যা ২০১৪ সালে ImageNet-এর টপ-৫ শ্রেণীবিন্যাসে ৯২.৭% সঠিকতা অর্জন করেছিল:
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি: [রিসার্চগেট](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493)
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
আমরা [Oxford-IIIT Pet Dataset](https://www.robots.ox.ac.uk/~vgg/data/pets/) ব্যবহার করব, যেখানে ৩৭টি ভিন্ন জাতের কুকুর এবং বিড়ালের ছবি রয়েছে।
|
||||
|
||||
|
||||
|
||||
|
||||
ডেটাসেট ডাউনলোড করতে, এই কোড স্নিপেটটি ব্যবহার করুন:
|
||||
|
||||
|
|
|
|||
|
|
@ -50,7 +50,7 @@
|
|||
"\n",
|
||||
"আদর্শ বিড়ালকে কল্পনা করার জন্য, আমরা একটি এলোমেলো শব্দযুক্ত ছবি দিয়ে শুরু করব এবং গ্রেডিয়েন্ট ডিসেন্ট অপ্টিমাইজেশন কৌশল ব্যবহার করে ছবিটি সামঞ্জস্য করার চেষ্টা করব, যাতে নেটওয়ার্কটি বিড়ালকে চিনতে পারে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"এখানে আমাদের শুরুর ছবি:\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -29,7 +29,7 @@ Keras এবং PyTorch-এ কিছু সাধারণ আর্কিট
|
|||
|
||||
এখানে VGG-16 নেটওয়ার্ক দ্বারা একটি বিড়ালের ছবির থেকে বের করা বৈশিষ্ট্যের উদাহরণ দেওয়া হয়েছে:
|
||||
|
||||
|
||||
|
||||
|
||||
## বিড়াল বনাম কুকুর ডেটাসেট
|
||||
|
||||
|
|
@ -48,19 +48,19 @@ Keras এবং PyTorch-এ কিছু সাধারণ আর্কিট
|
|||
|
||||
একটি পদ্ধতি হলো একটি র্যান্ডম ইমেজ দিয়ে শুরু করা এবং তারপর **গ্রেডিয়েন্ট ডিসেন্ট অপ্টিমাইজেশন** পদ্ধতি ব্যবহার করে সেই ইমেজটি এমনভাবে পরিবর্তন করা যাতে নেটওয়ার্কটি মনে করে এটি একটি বিড়াল।
|
||||
|
||||
|
||||
|
||||
|
||||
তবে, যদি আমরা এটি করি, তাহলে আমরা এমন কিছু পাব যা র্যান্ডম নয়েজের মতো দেখায়। কারণ *নেটওয়ার্ককে মনে করানোর অনেক উপায় আছে যে ইনপুট ইমেজটি একটি বিড়াল*, যার মধ্যে কিছু ভিজুয়ালি অর্থপূর্ণ নয়। যদিও এই ইমেজগুলোতে বিড়ালের জন্য সাধারণ প্যাটার্ন থাকে, তবুও এগুলোকে ভিজুয়ালি আলাদা করার জন্য কিছু নেই।
|
||||
|
||||
ফলাফল উন্নত করতে, আমরা লস ফাংশনে একটি নতুন টার্ম যোগ করতে পারি, যাকে **ভ্যারিয়েশন লস** বলা হয়। এটি একটি মেট্রিক যা দেখায় ইমেজের প্রতিবেশী পিক্সেলগুলো কতটা মিল। ভ্যারিয়েশন লস মিনিমাইজ করলে ইমেজটি মসৃণ হয় এবং নয়েজ দূর হয় - ফলে আরও ভিজুয়ালি আকর্ষণীয় প্যাটার্ন প্রকাশ পায়। এখানে এমন "আদর্শ" ইমেজের উদাহরণ দেওয়া হয়েছে, যা বিড়াল এবং জেব্রা হিসেবে উচ্চ সম্ভাবনায় শ্রেণীবদ্ধ করা হয়েছে:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*আদর্শ বিড়াল* | *আদর্শ জেব্রা*
|
||||
|
||||
একই পদ্ধতি ব্যবহার করে তথাকথিত **অ্যাডভার্সিয়াল আক্রমণ** করা যেতে পারে নিউরাল নেটওয়ার্কে। ধরুন আমরা একটি কুকুরকে বিড়াল হিসেবে দেখাতে চাই। যদি আমরা কুকুরের একটি ছবি নিই, যা নেটওয়ার্ক দ্বারা কুকুর হিসেবে স্বীকৃত, আমরা তারপর এটি সামান্য পরিবর্তন করতে পারি গ্রেডিয়েন্ট ডিসেন্ট অপ্টিমাইজেশন ব্যবহার করে, যতক্ষণ না নেটওয়ার্ক এটি বিড়াল হিসেবে শ্রেণীবদ্ধ করে:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*কুকুরের আসল ছবি* | *কুকুরের ছবি যা বিড়াল হিসেবে শ্রেণীবদ্ধ*
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@
|
|||
"\n",
|
||||
"যেহেতু আমরা অটোএনকোডারকে মূল ছবির যতটা সম্ভব তথ্য ধারণ করতে প্রশিক্ষণ দিচ্ছি সঠিক পুনর্গঠনের জন্য, নেটওয়ার্কটি ইনপুট ছবিগুলোর সেরা **এম্বেডিং** খুঁজে বের করার চেষ্টা করে অর্থ ধারণ করার জন্য।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> ছবি [Keras ব্লগ](https://blog.keras.io/building-autoencoders-in-keras.html) থেকে\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@
|
|||
"\n",
|
||||
"যেহেতু আমরা অটোএনকোডারকে মূল ছবির যতটা সম্ভব তথ্য ধারণ করতে প্রশিক্ষণ দিচ্ছি সঠিক পুনর্গঠনের জন্য, নেটওয়ার্কটি ইনপুট ছবিগুলোর সেরা **এম্বেডিং** খুঁজে বের করার চেষ্টা করে অর্থ ধারণ করার জন্য।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*ছবি [Keras ব্লগ](https://blog.keras.io/building-autoencoders-in-keras.html) থেকে*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ CNN প্রশিক্ষণের সময় একটি বড় সম
|
|||
|
||||
আমরা যখন অটোএনকোডার প্রশিক্ষণ দিই, তখন এটি মূল ইমেজ থেকে যতটা সম্ভব তথ্য ধারণ করার চেষ্টা করে, যাতে সঠিকভাবে পুনর্গঠন করা যায়। নেটওয়ার্কটি ইনপুট ইমেজগুলোর সেরা **এম্বেডিং** খুঁজে বের করার চেষ্টা করে, যা অর্থপূর্ণ তথ্য ধারণ করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [Keras ব্লগ](https://blog.keras.io/building-autoencoders-in-keras.html) থেকে
|
||||
|
||||
|
|
|
|||
|
|
@ -13,7 +13,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
## [পূর্ব-লেকচার কুইজ](https://ff-quizzes.netlify.app/en/ai/quiz/21)
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [YOLO v2 ওয়েবসাইট](https://pjreddie.com/darknet/yolov2/) থেকে নেওয়া
|
||||
|
||||
|
|
@ -25,7 +25,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
2. প্রতিটি টাইলের উপর ইমেজ ক্লাসিফিকেশন চালান।
|
||||
3. যেসব টাইল যথেষ্ট উচ্চ অ্যাক্টিভেশন দেখায়, সেগুলোতে কাঙ্ক্ষিত বস্তুটি রয়েছে বলে বিবেচনা করুন।
|
||||
|
||||
|
||||
|
||||
|
||||
> *ছবি [এক্সারসাইজ নোটবুক](ObjectDetection-TF.ipynb) থেকে নেওয়া*
|
||||
|
||||
|
|
@ -42,7 +42,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
* [PASCAL VOC](http://host.robots.ox.ac.uk/pascal/VOC/) - ২০টি ক্লাস
|
||||
* [COCO](http://cocodataset.org/#home) - সাধারণ বস্তুসমূহের প্রসঙ্গ। ৮০টি ক্লাস, বাউন্ডিং বক্স এবং সেগমেন্টেশন মাস্ক।
|
||||
|
||||
|
||||
|
||||
|
||||
## অবজেক্ট ডিটেকশন মেট্রিকস
|
||||
|
||||
|
|
@ -50,7 +50,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
যেখানে ইমেজ ক্লাসিফিকেশনের ক্ষেত্রে অ্যালগরিদম কতটা ভালো কাজ করছে তা পরিমাপ করা সহজ, অবজেক্ট ডিটেকশনের ক্ষেত্রে আমাদের ক্লাসের সঠিকতা এবং অনুমিত বাউন্ডিং বক্সের অবস্থানের নির্ভুলতা উভয়ই পরিমাপ করতে হবে। এর জন্য আমরা **ইন্টারসেকশন ওভার ইউনিয়ন** (IoU) ব্যবহার করি, যা দুটি বক্স (বা দুটি এলাকা) কতটা ওভারল্যাপ করছে তা পরিমাপ করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> *[এই চমৎকার ব্লগ পোস্ট](https://pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/) থেকে ফিগার ২*
|
||||
|
||||
|
|
@ -98,11 +98,11 @@ $$
|
|||
|
||||
[R-CNN](http://islab.ulsan.ac.kr/files/announcement/513/rcnn_pami.pdf) [Selective Search](http://www.huppelen.nl/publications/selectiveSearchDraft.pdf) ব্যবহার করে ROI অঞ্চলের একটি হায়ারারকিকাল স্ট্রাকচার তৈরি করে, যা পরে CNN ফিচার এক্সট্রাক্টর এবং SVM-ক্লাসিফায়ারগুলোর মাধ্যমে বস্তু ক্লাস নির্ধারণ করে এবং লিনিয়ার রিগ্রেশন ব্যবহার করে *বাউন্ডিং বক্স* এর কোঅর্ডিনেট নির্ধারণ করে। [অফিশিয়াল পেপার](https://arxiv.org/pdf/1506.01497v1.pdf)
|
||||
|
||||
|
||||
|
||||
|
||||
> *ছবি van de Sande et al. ICCV’11*
|
||||
|
||||
|
||||
|
||||
|
||||
> *ছবি [এই ব্লগ](https://towardsdatascience.com/r-cnn-fast-r-cnn-faster-r-cnn-yolo-object-detection-algorithms-36d53571365e) থেকে নেওয়া*
|
||||
|
||||
|
|
@ -110,7 +110,7 @@ $$
|
|||
|
||||
এই পদ্ধতি R-CNN এর মতোই, তবে রিজনগুলো কনভোলিউশন লেয়ার প্রয়োগ করার পরে নির্ধারণ করা হয়।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [অফিশিয়াল পেপার](https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Girshick_Fast_R-CNN_ICCV_2015_paper.pdf), [arXiv](https://arxiv.org/pdf/1504.08083.pdf), ২০১৫
|
||||
|
||||
|
|
@ -118,7 +118,7 @@ $$
|
|||
|
||||
এই পদ্ধতির মূল ধারণাটি হল রিজন প্রেডিক্ট করতে নিউরাল নেটওয়ার্ক ব্যবহার করা - যাকে *রিজন প্রপোজাল নেটওয়ার্ক* বলা হয়। [পেপার](https://arxiv.org/pdf/1506.01497.pdf), ২০১৬
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [অফিশিয়াল পেপার](https://arxiv.org/pdf/1506.01497.pdf) থেকে নেওয়া
|
||||
|
||||
|
|
@ -130,7 +130,7 @@ $$
|
|||
2. ফিচারগুলো **পজিশন-সেনসিটিভ স্কোর ম্যাপ** দ্বারা প্রক্রিয়াকৃত হয়। $C$ ক্লাসের প্রতিটি অবজেক্টকে $k\times k$ অঞ্চলে ভাগ করা হয় এবং আমরা অবজেক্টের অংশগুলো প্রেডিক্ট করতে প্রশিক্ষণ দিই।
|
||||
3. $k\times k$ অঞ্চলের প্রতিটি অংশের জন্য সমস্ত নেটওয়ার্ক অবজেক্ট ক্লাসের জন্য ভোট দেয় এবং সর্বাধিক ভোট পাওয়া অবজেক্ট ক্লাসটি নির্বাচিত হয়।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [অফিশিয়াল পেপার](https://arxiv.org/abs/1605.06409) থেকে নেওয়া
|
||||
|
||||
|
|
@ -141,7 +141,7 @@ YOLO একটি রিয়েলটাইম ওয়ান-পাস অ
|
|||
* ছবিটিকে $S\times S$ অঞ্চলে ভাগ করা হয়।
|
||||
* প্রতিটি অঞ্চলের জন্য, **CNN** $n$ সম্ভাব্য অবজেক্ট, *বাউন্ডিং বক্স* এর কোঅর্ডিনেট এবং *কনফিডেন্স*=*প্রোবাবিলিটি* * IoU প্রেডিক্ট করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [অফিশিয়াল পেপার](https://arxiv.org/abs/1506.02640) থেকে নেওয়া
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# কম্পিউটার ভিশন
|
||||
|
||||
|
||||
|
||||
|
||||
এই অংশে আমরা শিখব:
|
||||
|
||||
|
|
|
|||
|
|
@ -199,7 +199,7 @@
|
|||
"\n",
|
||||
"**শব্দের ব্যাগ** (BoW) ভেক্টর উপস্থাপনাটি সবচেয়ে বেশি ব্যবহৃত ঐতিহ্যবাহী ভেক্টর উপস্থাপনাগুলোর একটি। প্রতিটি শব্দ একটি ভেক্টর সূচকের সাথে যুক্ত থাকে, এবং ভেক্টরের উপাদানটি একটি নির্দিষ্ট ডকুমেন্টে সেই শব্দটির উপস্থিতির সংখ্যা ধারণ করে।\n",
|
||||
"\n",
|
||||
" \n",
|
||||
" \n",
|
||||
"\n",
|
||||
"> **Note**: BoW-কে আপনি টেক্সটের পৃথক শব্দগুলোর জন্য এক-হট-এনকোডেড ভেক্টরগুলোর যোগফল হিসেবেও ভাবতে পারেন।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -188,7 +188,7 @@
|
|||
"\n",
|
||||
"**ব্যাগ-অফ-ওয়ার্ডস** (BoW) ভেক্টর উপস্থাপন হল সবচেয়ে সহজে বোঝা যায় এমন ঐতিহ্যবাহী ভেক্টর উপস্থাপন। প্রতিটি শব্দ একটি ভেক্টর সূচকের সাথে যুক্ত থাকে, এবং একটি ভেক্টর উপাদান একটি নির্দিষ্ট ডকুমেন্টে প্রতিটি শব্দের উপস্থিতির সংখ্যা ধারণ করে।\n",
|
||||
"\n",
|
||||
" \n",
|
||||
" \n",
|
||||
"\n",
|
||||
"> **Note**: BoW কে আপনি টেক্সটের পৃথক শব্দগুলোর জন্য সমস্ত এক-হট-এনকোডেড ভেক্টরের যোগফল হিসেবেও ভাবতে পারেন।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -62,7 +62,7 @@
|
|||
"\n",
|
||||
"আমাদের নেটওয়ার্কে প্রথম লেয়ার হিসেবে এমবেডিং লেয়ার ব্যবহার করে, আমরা ব্যাগ-অফ-ওয়ার্ডস থেকে **এমবেডিং ব্যাগ** মডেলে পরিবর্তন করতে পারি, যেখানে আমরা প্রথমে আমাদের টেক্সটের প্রতিটি শব্দকে সংশ্লিষ্ট এমবেডিংয়ে রূপান্তর করি এবং তারপর সেই সমস্ত এমবেডিংয়ের উপর কিছু সামগ্রিক ফাংশন গণনা করি, যেমন `sum`, `average` বা `max`।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"আমাদের ক্লাসিফায়ার নিউরাল নেটওয়ার্কটি এমবেডিং লেয়ার দিয়ে শুরু হবে, তারপর অ্যাগ্রিগেশন লেয়ার, এবং তার উপরে একটি লিনিয়ার ক্লাসিফায়ার:\n"
|
||||
]
|
||||
|
|
@ -176,7 +176,7 @@
|
|||
"\n",
|
||||
"পূর্ববর্তী আর্কিটেকচারে, সমস্ত সিকোয়েন্সকে একই দৈর্ঘ্যে প্যাড করতে হতো যাতে সেগুলোকে একটি মিনিব্যাচে ফিট করানো যায়। এটি ভেরিয়েবল দৈর্ঘ্যের সিকোয়েন্স উপস্থাপনার সবচেয়ে কার্যকর পদ্ধতি নয় - আরেকটি পদ্ধতি হতে পারে **অফসেট** ভেক্টর ব্যবহার করা, যা একটি বড় ভেক্টরে সংরক্ষিত সমস্ত সিকোয়েন্সের অফসেট ধারণ করবে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> **Note**: উপরের ছবিতে, আমরা একটি ক্যারেক্টারের সিকোয়েন্স দেখিয়েছি, কিন্তু আমাদের উদাহরণে আমরা শব্দের সিকোয়েন্স নিয়ে কাজ করছি। তবে, অফসেট ভেক্টর দিয়ে সিকোয়েন্স উপস্থাপনার সাধারণ নীতিটি একই থাকে।\n",
|
||||
"\n",
|
||||
|
|
@ -311,7 +311,7 @@
|
|||
"\n",
|
||||
"CBoW দ্রুততর, তবে স্কিপ-গ্রাম ধীর হলেও কম ঘন ঘন ব্যবহৃত শব্দগুলোর উপস্থাপনা করতে ভালো কাজ করে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Google News ডেটাসেটে প্রি-ট্রেন করা word2vec এম্বেডিং নিয়ে পরীক্ষা করার জন্য, আমরা **gensim** লাইব্রেরি ব্যবহার করতে পারি। নিচে 'neural' শব্দটির সাথে সবচেয়ে মিল থাকা শব্দগুলো খুঁজে বের করা হয়েছে:\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -37,7 +37,7 @@
|
|||
"\n",
|
||||
"আমাদের নেটওয়ার্কে প্রথম লেয়ার হিসেবে একটি এম্বেডিং লেয়ার ব্যবহার করে, আমরা ব্যাগ-অফ-ওয়ার্ডস থেকে **এম্বেডিং ব্যাগ** মডেলে স্যুইচ করতে পারি, যেখানে আমরা প্রথমে আমাদের টেক্সটের প্রতিটি শব্দকে সংশ্লিষ্ট এম্বেডিং-এ রূপান্তর করি এবং তারপর সেই এম্বেডিংগুলোর উপর একটি সমষ্টিগত ফাংশন (যেমন `sum`, `average` বা `max`) গণনা করি।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"আমাদের ক্লাসিফায়ার নিউরাল নেটওয়ার্কে নিম্নলিখিত লেয়ারগুলো অন্তর্ভুক্ত রয়েছে:\n",
|
||||
"\n",
|
||||
|
|
@ -283,7 +283,7 @@
|
|||
"\n",
|
||||
"CBoW দ্রুততর, এবং স্কিপ-গ্রাম ধীর হলেও এটি কম ঘন ঘন ব্যবহৃত শব্দগুলোর উপস্থাপনা আরও ভালোভাবে করে।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"গুগল নিউজ ডেটাসেটে প্রি-ট্রেইন করা Word2Vec এম্বেডিং নিয়ে পরীক্ষা করার জন্য, আমরা **gensim** লাইব্রেরি ব্যবহার করতে পারি। নিচে আমরা 'neural' শব্দটির সাথে সবচেয়ে সাদৃশ্যপূর্ণ শব্দগুলো খুঁজে বের করব।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ BoW বা TF/IDF ভিত্তিক ক্লাসিফায়ার প
|
|||
|
||||
আমাদের ক্লাসিফায়ার নেটওয়ার্কে প্রথম লেয়ার হিসেবে একটি এমবেডিং লেয়ার ব্যবহার করে, আমরা ব্যাগ-অফ-ওয়ার্ডস থেকে **এমবেডিং ব্যাগ** মডেলে পরিবর্তন করতে পারি, যেখানে আমরা প্রথমে আমাদের টেক্সটের প্রতিটি শব্দকে সংশ্লিষ্ট এমবেডিংয়ে রূপান্তর করি এবং তারপর সেই এমবেডিংগুলোর উপর কিছু অ্যাগ্রিগেট ফাংশন গণনা করি, যেমন `sum`, `average` বা `max`।
|
||||
|
||||
|
||||
|
||||
|
||||
> লেখকের তৈরি চিত্র
|
||||
|
||||
|
|
@ -40,7 +40,7 @@ BoW বা TF/IDF ভিত্তিক ক্লাসিফায়ার প
|
|||
|
||||
CBoW দ্রুততর, তবে স্কিপ-গ্রাম ধীরতর, কিন্তু কম ঘন ঘন ব্যবহৃত শব্দগুলোকে ভালোভাবে রিপ্রেজেন্ট করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> [এই পেপার](https://arxiv.org/pdf/1301.3781.pdf) থেকে নেওয়া চিত্র
|
||||
|
||||
|
|
|
|||
|
|
@ -23,7 +23,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
* **কন্টিনিউয়াস ব্যাগ-অফ-ওয়ার্ডস** (CBoW), যেখানে আমরা একটি টোকেন সিকোয়েন্স $W_{-N}$, ..., $W_N$ এর মধ্যে $W_0$ টোকেন পূর্বানুমান করি।
|
||||
* **স্কিপ-গ্রাম**, যেখানে আমরা মধ্যবর্তী টোকেন $W_0$ থেকে পার্শ্ববর্তী টোকেনগুলোর সেট {$W_{-N},\dots, W_{-1}, W_1,\dots, W_N$} পূর্বানুমান করি।
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি [এই পেপার](https://arxiv.org/pdf/1301.3781.pdf) থেকে
|
||||
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
টেক্সট সিকোয়েন্সের অর্থ ধারণ করতে, আমাদের আরেকটি নিউরাল নেটওয়ার্ক আর্কিটেকচার ব্যবহার করতে হবে, যেটিকে **রিকারেন্ট নিউরাল নেটওয়ার্ক** বা RNN বলা হয়। RNN-এ, আমরা আমাদের বাক্যটি নেটওয়ার্কের মাধ্যমে একবারে একটি প্রতীক পাঠাই, এবং নেটওয়ার্ক কিছু **স্টেট** তৈরি করে, যা আমরা পরবর্তী প্রতীকের সাথে আবার নেটওয়ার্কে পাঠাই।
|
||||
|
||||
|
||||
|
||||
|
||||
> লেখকের তৈরি ছবি
|
||||
|
||||
|
|
@ -61,7 +61,7 @@ LSTM নেটওয়ার্ক RNN-এর মতোই সংগঠিত,
|
|||
|
||||
একটি রিকারেন্ট নেটওয়ার্ক, একদিকে বা বাইডিরেকশনাল, একটি সিকোয়েন্সের নির্দিষ্ট প্যাটার্নগুলো ধারণ করে এবং সেগুলোকে স্টেট ভেক্টরে সংরক্ষণ করতে পারে বা আউটপুটে পাস করতে পারে। কনভোলিউশনাল নেটওয়ার্কের মতো, আমরা প্রথম লেয়ার দ্বারা এক্সট্র্যাক্ট করা নিম্ন-স্তরের প্যাটার্ন থেকে উচ্চ-স্তরের প্যাটার্ন ক্যাপচার করতে প্রথম লেয়ারের উপরে আরেকটি রিকারেন্ট লেয়ার তৈরি করতে পারি। এটি আমাদের **মাল্টিলেয়ার RNN** ধারণায় নিয়ে যায়, যা দুটি বা তার বেশি রিকারেন্ট নেটওয়ার্ক নিয়ে গঠিত, যেখানে পূর্ববর্তী লেয়ারের আউটপুট পরবর্তী লেয়ারের ইনপুট হিসেবে পাঠানো হয়।
|
||||
|
||||
|
||||
|
||||
|
||||
*ছবি [এই অসাধারণ পোস্ট](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) থেকে ফার্নান্দো লোপেজের দ্বারা*
|
||||
|
||||
|
|
|
|||
|
|
@ -422,7 +422,7 @@
|
|||
"\n",
|
||||
"পুনরাবৃত্ত নেটওয়ার্ক, একমুখী বা দ্বিমুখী, একটি সিকোয়েন্সের নির্দিষ্ট প্যাটার্নগুলো ধারণ করে এবং সেগুলো স্টেট ভেক্টরে সংরক্ষণ করতে পারে বা আউটপুটে পাস করতে পারে। কনভোলিউশনাল নেটওয়ার্কের মতো, আমরা প্রথম স্তরের উপর আরেকটি পুনরাবৃত্ত স্তর তৈরি করতে পারি, যা নিম্ন-স্তরের প্যাটার্ন থেকে উচ্চ-স্তরের প্যাটার্ন ধারণ করে। এটি আমাদের **বহুস্তর RNN** ধারণার দিকে নিয়ে যায়, যা দুই বা তার বেশি পুনরাবৃত্ত নেটওয়ার্ক নিয়ে গঠিত, যেখানে পূর্ববর্তী স্তরের আউটপুট পরবর্তী স্তরের ইনপুট হিসেবে পাস করা হয়।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*ছবি [এই অসাধারণ পোস্ট](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) থেকে, লেখক Fernando López*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
"\n",
|
||||
"একটি টেক্সট সিকোয়েন্সের অর্থ ধরতে, আমরা **পুনরাবৃত্তি নিউরাল নেটওয়ার্ক** বা RNN নামে একটি নিউরাল নেটওয়ার্ক আর্কিটেকচার ব্যবহার করব। RNN ব্যবহার করার সময়, আমরা আমাদের বাক্যটি নেটওয়ার্কের মধ্য দিয়ে একবারে একটি টোকেন পাঠাই, এবং নেটওয়ার্কটি কিছু **অবস্থা** তৈরি করে, যা আমরা পরবর্তী টোকেনের সাথে আবার নেটওয়ার্কে পাঠাই।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"টোকেনগুলির ইনপুট সিকোয়েন্স $X_0,\\dots,X_n$ দেওয়া হলে, RNN একটি নিউরাল নেটওয়ার্ক ব্লকের সিকোয়েন্স তৈরি করে এবং ব্যাকপ্রোপাগেশনের মাধ্যমে এই সিকোয়েন্সটি এন্ড-টু-এন্ড প্রশিক্ষণ দেয়। প্রতিটি নেটওয়ার্ক ব্লক $(X_i,S_i)$ জোড়াকে ইনপুট হিসাবে নেয় এবং ফলাফল হিসাবে $S_{i+1}$ তৈরি করে। চূড়ান্ত অবস্থা $S_n$ বা আউটপুট $Y_n$ একটি লিনিয়ার ক্লাসিফায়ারে যায় ফলাফল তৈরি করতে। সমস্ত নেটওয়ার্ক ব্লক একই ওজন ভাগ করে এবং একটি ব্যাকপ্রোপাগেশন পাস ব্যবহার করে এন্ড-টু-এন্ড প্রশিক্ষণ দেওয়া হয়।\n",
|
||||
"\n",
|
||||
|
|
@ -371,7 +371,7 @@
|
|||
"\n",
|
||||
"পুনরাবৃত্ত নেটওয়ার্ক, একমুখী বা দ্বিমুখী, সিকোয়েন্সের মধ্যে প্যাটার্নগুলো ধরে এবং সেগুলোকে স্টেট ভেক্টরে সংরক্ষণ করে বা আউটপুট হিসেবে ফেরত দেয়। কনভোলিউশনাল নেটওয়ার্কের মতো, আমরা প্রথম স্তরের মাধ্যমে নিম্ন স্তরের প্যাটার্নগুলো থেকে উচ্চ স্তরের প্যাটার্ন ধরার জন্য আরেকটি পুনরাবৃত্ত স্তর তৈরি করতে পারি। এটি আমাদের **বহুস্তর RNN** ধারণার দিকে নিয়ে যায়, যা দুটি বা তার বেশি পুনরাবৃত্ত নেটওয়ার্ক নিয়ে গঠিত, যেখানে আগের স্তরের আউটপুট পরবর্তী স্তরের ইনপুট হিসেবে ব্যবহৃত হয়।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*ছবি [এই চমৎকার পোস্ট](https://towardsdatascience.com/from-a-lstm-cell-to-a-multilayer-lstm-network-with-pytorch-2899eb5696f3) থেকে, লেখক Fernando López।*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -119,7 +119,7 @@
|
|||
"\n",
|
||||
"আমরা RNN-কে টেক্সট জেনারেট করার জন্য যেভাবে প্রশিক্ষণ দেব তা হলো নিম্নরূপ। প্রতিটি ধাপে, আমরা `nchars` দৈর্ঘ্যের একটি অক্ষরের ক্রম নেব এবং নেটওয়ার্ককে প্রতিটি ইনপুট অক্ষরের জন্য পরবর্তী আউটপুট অক্ষর তৈরি করতে বলব:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"বাস্তব পরিস্থিতির উপর নির্ভর করে, আমরা কিছু বিশেষ অক্ষর অন্তর্ভুক্ত করতে চাইতে পারি, যেমন *end-of-sequence* `<eos>`। আমাদের ক্ষেত্রে, আমরা শুধু নেটওয়ার্ককে অবিরাম টেক্সট জেনারেশনের জন্য প্রশিক্ষণ দিতে চাই, তাই আমরা প্রতিটি ক্রমের আকারকে `nchars` টোকেনের সমান স্থির করব। ফলস্বরূপ, প্রতিটি প্রশিক্ষণ উদাহরণে থাকবে `nchars` ইনপুট এবং `nchars` আউটপুট (যা ইনপুট ক্রম এক প্রতীক বামে সরানো)। মিনিব্যাচে এমন কয়েকটি ক্রম থাকবে।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -113,7 +113,7 @@
|
|||
"\n",
|
||||
"আমরা যেভাবে RNN প্রশিক্ষণ করব যাতে এটি সংবাদ শিরোনাম তৈরি করতে পারে তা হলো নিম্নরূপ। প্রতিটি ধাপে, আমরা একটি শিরোনাম নেব, যা RNN-এ প্রবেশ করানো হবে, এবং প্রতিটি ইনপুট অক্ষরের জন্য আমরা নেটওয়ার্ককে পরবর্তী আউটপুট অক্ষর তৈরি করতে বলব:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"আমাদের ক্রমের শেষ অক্ষরের জন্য, আমরা নেটওয়ার্ককে `<eos>` টোকেন তৈরি করতে বলব।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ Recurrent Neural Networks (RNNs) এবং তাদের গেটেড স
|
|||
|
||||
এটি বিভিন্ন নিউরাল আর্কিটেকচারের জন্য অনুমতি দেয়, যা নিচের ছবিতে দেখানো হয়েছে:
|
||||
|
||||
|
||||
|
||||
|
||||
> ছবি ব্লগ পোস্ট [Unreasonable Effectiveness of Recurrent Neural Networks](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) থেকে [Andrej Karpaty](http://karpathy.github.io/) দ্বারা
|
||||
|
||||
|
|
@ -32,7 +32,7 @@ Recurrent Neural Networks (RNNs) এবং তাদের গেটেড স
|
|||
|
||||
আমরা এই RNN-কে ধাপে ধাপে টেক্সট তৈরি করতে প্রশিক্ষণ দেব। প্রতিটি ধাপে, আমরা `nchars` দৈর্ঘ্যের একটি চরিত্রের সিকোয়েন্স গ্রহণ করব এবং নেটওয়ার্ককে প্রতিটি ইনপুট চরিত্রের জন্য পরবর্তী আউটপুট চরিত্র তৈরি করতে বলব:
|
||||
|
||||
|
||||
|
||||
|
||||
যখন টেক্সট তৈরি করা হয় (ইনফারেন্সের সময়), আমরা কিছু **প্রম্পট** দিয়ে শুরু করি, যা RNN সেলগুলোর মাধ্যমে পাস করে তার মধ্যবর্তী স্টেট তৈরি করে, এবং তারপর এই স্টেট থেকে জেনারেশন শুরু হয়। আমরা একবারে একটি চরিত্র তৈরি করি এবং স্টেট এবং তৈরি করা চরিত্রকে অন্য RNN সেলে পাস করি পরবর্তী চরিত্র তৈরি করার জন্য, যতক্ষণ না আমরা পর্যাপ্ত চরিত্র তৈরি করি।
|
||||
|
||||
|
|
|
|||
|
|
@ -20,13 +20,13 @@ RNN ব্যবহার করে সিকোয়েন্স-টু-সি
|
|||
|
||||
**অ্যাটেনশন মেকানিজম** RNN-এর আউটপুট প্রেডিকশনে প্রতিটি ইনপুট ভেক্টরের প্রসঙ্গগত প্রভাবকে ওজন করার একটি উপায় প্রদান করে। এটি বাস্তবায়িত হয় ইনপুট RNN এবং আউটপুট RNN-এর মধ্যবর্তী স্টেটগুলোর মধ্যে শর্টকাট তৈরি করে। এইভাবে, আউটপুট প্রতীক y<sub>t</sub> তৈরি করার সময়, আমরা সমস্ত ইনপুট হিডেন স্টেট h<sub>i</sub> বিবেচনা করব, বিভিন্ন ওজন সহগ α<sub>t,i</sub> সহ।
|
||||
|
||||
|
||||
|
||||
|
||||
> [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf)-এ অ্যাডিটিভ অ্যাটেনশন মেকানিজম সহ এনকোডার-ডিকোডার মডেল, [এই ব্লগ পোস্ট](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html) থেকে উদ্ধৃত।
|
||||
|
||||
অ্যাটেনশন ম্যাট্রিক্স {α<sub>i,j</sub>} একটি আউটপুট সিকোয়েন্সের একটি নির্দিষ্ট শব্দ তৈরিতে ইনপুট শব্দগুলো কীভাবে ভূমিকা রাখে তা উপস্থাপন করবে। নিচে এমন একটি ম্যাট্রিক্সের উদাহরণ দেওয়া হয়েছে:
|
||||
|
||||
|
||||
|
||||
|
||||
> [Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) থেকে চিত্র (Fig.3)
|
||||
|
||||
|
|
@ -66,7 +66,7 @@ RNN ব্যবহার করে সিকোয়েন্স-টু-সি
|
|||
|
||||
এরপর, আমাদের সিকোয়েন্সের মধ্যে কিছু প্যাটার্ন ধরতে হবে। এটি করতে, ট্রান্সফর্মারগুলো **সেলফ-অ্যাটেনশন** মেকানিজম ব্যবহার করে, যা মূলত ইনপুট এবং আউটপুট হিসেবে একই সিকোয়েন্সে অ্যাটেনশন প্রয়োগ। সেলফ-অ্যাটেনশন প্রয়োগ করে আমরা বাক্যের প্রসঙ্গ বিবেচনা করতে পারি এবং কোন শব্দগুলো আন্তঃসম্পর্কিত তা দেখতে পারি। উদাহরণস্বরূপ, এটি আমাদের *it* দ্বারা উল্লেখিত শব্দগুলো দেখতে এবং প্রসঙ্গ বিবেচনা করতে সাহায্য করে:
|
||||
|
||||
|
||||
|
||||
|
||||
> [গুগলের ব্লগ](https://research.googleblog.com/2017/08/transformer-novel-neural-network.html) থেকে চিত্র।
|
||||
|
||||
|
|
@ -91,7 +91,7 @@ RNN ব্যবহার করে সিকোয়েন্স-টু-সি
|
|||
|
||||
**BERT** (Bidirectional Encoder Representations from Transformers) একটি খুব বড় মাল্টি লেয়ার ট্রান্সফর্মার নেটওয়ার্ক, যেখানে *BERT-base*-এর জন্য 12টি লেয়ার এবং *BERT-large*-এর জন্য 24টি লেয়ার রয়েছে। মডেলটি প্রথমে একটি বড় টেক্সট ডেটা কর্পাস (WikiPedia + বই) ব্যবহার করে আনসুপারভাইজড প্রশিক্ষণের মাধ্যমে প্রি-ট্রেইন করা হয় (একটি বাক্যে মাস্ক করা শব্দগুলো প্রেডিক্ট করা)। প্রি-ট্রেইনিংয়ের সময় মডেলটি উল্লেখযোগ্য ভাষা বোঝার স্তর অর্জন করে, যা পরে অন্যান্য ডেটাসেটের সাথে ফাইন টিউনিংয়ের মাধ্যমে ব্যবহার করা যায়। এই প্রক্রিয়াকে **ট্রান্সফার লার্নিং** বলা হয়।
|
||||
|
||||
|
||||
|
||||
|
||||
> [উৎস](http://jalammar.github.io/illustrated-bert/)
|
||||
|
||||
|
|
|
|||
|
|
@ -12,12 +12,12 @@
|
|||
"\n",
|
||||
"**অ্যাটেনশন মেকানিজম** একটি পদ্ধতি প্রদান করে, যা প্রতিটি ইনপুট ভেক্টরের প্রাসঙ্গিক প্রভাবকে RNN-এর প্রতিটি আউটপুট প্রেডিকশনে ওজন দেয়। এটি বাস্তবায়িত হয় ইনপুট RNN-এর মধ্যবর্তী স্টেট এবং আউটপুট RNN-এর মধ্যে শর্টকাট তৈরি করে। এই পদ্ধতিতে, যখন আউটপুট প্রতীক $y_t$ তৈরি করা হয়, তখন আমরা সব ইনপুট হিডেন স্টেট $h_i$-কে বিভিন্ন ওজন সহগ $\\alpha_{t,i}$ সহ বিবেচনা করব।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*[Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) থেকে উদ্ধৃত, [এই ব্লগ পোস্ট](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html) থেকে নেওয়া এনকোডার-ডিকোডার মডেল অ্যাডিটিভ অ্যাটেনশন মেকানিজম সহ]*\n",
|
||||
"\n",
|
||||
"অ্যাটেনশন ম্যাট্রিক্স $\\{\\alpha_{i,j}\\}$ নির্দেশ করে যে নির্দিষ্ট ইনপুট শব্দগুলো আউটপুট সিকোয়েন্সের একটি নির্দিষ্ট শব্দ তৈরিতে কতটা ভূমিকা রাখে। নিচে এমন একটি ম্যাট্রিক্সের উদাহরণ দেওয়া হলো:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*[Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) থেকে নেওয়া (Fig.3)*\n",
|
||||
"\n",
|
||||
|
|
@ -35,7 +35,7 @@
|
|||
"\n",
|
||||
"**BERT** (Bidirectional Encoder Representations from Transformers) একটি খুব বড় মাল্টি-লেয়ার ট্রান্সফর্মার নেটওয়ার্ক, যেখানে *BERT-base*-এর জন্য ১২টি লেয়ার এবং *BERT-large*-এর জন্য ২৪টি লেয়ার রয়েছে। মডেলটি প্রথমে একটি বড় টেক্সট ডেটাসেট (উইকিপিডিয়া + বই) ব্যবহার করে আনসুপারভাইজড প্রশিক্ষণের মাধ্যমে প্রি-ট্রেইন করা হয় (একটি বাক্যে মাস্ক করা শব্দগুলো প্রেডিক্ট করা)। প্রি-ট্রেইনিংয়ের সময় মডেলটি উল্লেখযোগ্য ভাষাগত বোঝাপড়া অর্জন করে, যা পরে অন্যান্য ডেটাসেটের সাথে ফাইন টিউনিংয়ের মাধ্যমে ব্যবহার করা যায়। এই প্রক্রিয়াটিকে **ট্রান্সফার লার্নিং** বলা হয়।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"ট্রান্সফর্মার আর্কিটেকচারের অনেক ভেরিয়েশন রয়েছে, যেমন BERT, DistilBERT, BigBird, OpenGPT3 এবং আরও অনেক, যেগুলো ফাইন টিউন করা যায়। [HuggingFace প্যাকেজ](https://github.com/huggingface/) PyTorch ব্যবহার করে এই আর্কিটেকচারগুলোর অনেকগুলোর প্রশিক্ষণের জন্য একটি রিপোজিটরি প্রদান করে।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -12,12 +12,12 @@
|
|||
"\n",
|
||||
"**অ্যাটেনশন মেকানিজম** RNN-এর প্রতিটি আউটপুট প্রেডিকশনে প্রতিটি ইনপুট ভেক্টরের প্রাসঙ্গিক প্রভাবকে ওজন দেওয়ার একটি উপায় প্রদান করে। এটি বাস্তবায়িত হয় ইনপুট RNN-এর মধ্যবর্তী স্টেট এবং আউটপুট RNN-এর মধ্যে শর্টকাট তৈরি করে। এই পদ্ধতিতে, যখন আউটপুট প্রতীক $y_t$ তৈরি করা হয়, তখন আমরা বিভিন্ন ওজন সহগ $\\alpha_{t,i}$ সহ সমস্ত ইনপুট হিডেন স্টেট $h_i$ বিবেচনায় নেব। \n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*[Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) থেকে উদ্ধৃত, [এই ব্লগ পোস্ট](https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html) থেকে নেওয়া এনকোডার-ডিকোডার মডেল অ্যাডিটিভ অ্যাটেনশন মেকানিজম সহ*\n",
|
||||
"\n",
|
||||
"অ্যাটেনশন ম্যাট্রিক্স $\\{\\alpha_{i,j}\\}$ একটি নির্দিষ্ট আউটপুট সিকোয়েন্সের একটি শব্দ তৈরিতে কোন ইনপুট শব্দগুলো কতটা ভূমিকা রাখছে তা উপস্থাপন করে। নিচে এমন একটি ম্যাট্রিক্সের উদাহরণ দেওয়া হলো:\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*[Bahdanau et al., 2015](https://arxiv.org/pdf/1409.0473.pdf) থেকে নেওয়া (Fig.3)*\n",
|
||||
"\n",
|
||||
|
|
@ -225,7 +225,7 @@
|
|||
"\n",
|
||||
"**BERT** (Bidirectional Encoder Representations from Transformers) একটি অত্যন্ত বড় মাল্টি-লেয়ার ট্রান্সফর্মার নেটওয়ার্ক, যেখানে *BERT-base* এর জন্য ১২টি স্তর এবং *BERT-large* এর জন্য ২৪টি স্তর রয়েছে। এই মডেলটি প্রথমে বিশাল পরিমাণ টেক্সট ডেটা (উইকিপিডিয়া + বই) ব্যবহার করে অ-পর্যবেক্ষণমূলক প্রশিক্ষণের মাধ্যমে (একটি বাক্যে মাস্ক করা শব্দ অনুমান করা) প্রি-ট্রেইন করা হয়। প্রি-ট্রেইনিংয়ের সময় মডেলটি উল্লেখযোগ্য ভাষাগত বোঝাপড়া অর্জন করে, যা পরে অন্যান্য ডেটাসেটের সাথে ফাইন টিউনিংয়ের মাধ্যমে ব্যবহার করা যায়। এই প্রক্রিয়াকে **ট্রান্সফার লার্নিং** বলা হয়।\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"ট্রান্সফর্মার আর্কিটেকচারের অনেক বৈচিত্র্য রয়েছে, যেমন BERT, DistilBERT, BigBird, OpenGPT3 এবং আরও অনেক কিছু, যেগুলো ফাইন টিউন করা যেতে পারে।\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ infant | O
|
|||
|
||||
যেহেতু আমাদের টোকেন এবং শ্রেণীগুলোর মধ্যে এক-একটি সম্পর্ক তৈরি করতে হবে, আমরা এই চিত্র থেকে একটি ডানদিকের **অনেক-থেকে-অনেক** নিউরাল নেটওয়ার্ক মডেল প্রশিক্ষণ দিতে পারি:
|
||||
|
||||
|
||||
|
||||
|
||||
> *[এই ব্লগ পোস্ট](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) থেকে চিত্র, লেখক [Andrej Karpathy](http://karpathy.github.io/)। NER টোকেন শ্রেণীবিন্যাস মডেল এই চিত্রের ডানদিকের নেটওয়ার্ক আর্কিটেকচারের সাথে মিলে যায়।*
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# প্রাকৃতিক ভাষা প্রক্রিয়াকরণ
|
||||
|
||||
|
||||
|
||||
|
||||
এই অংশে, আমরা **প্রাকৃতিক ভাষা প্রক্রিয়াকরণ (NLP)** সম্পর্কিত কাজগুলো পরিচালনা করতে নিউরাল নেটওয়ার্ক ব্যবহার করার উপর মনোযোগ দেব। অনেক ধরনের NLP সমস্যা রয়েছে যা আমরা চাই কম্পিউটার সমাধান করতে সক্ষম হোক:
|
||||
|
||||
|
|
|
|||
|
|
@ -70,7 +70,7 @@ NetLogo-এর একটি চমৎকার দিক হলো এটি এ
|
|||
|
||||
মডেলটি খোলার পরে, আপনি NetLogo-এর প্রধান স্ক্রিনে নিয়ে যাওয়া হবে। এখানে একটি নমুনা মডেল রয়েছে যা সীমিত সম্পদ (ঘাস) দেওয়া হলে নেকড়ে এবং ভেড়ার জনসংখ্যা বর্ণনা করে।
|
||||
|
||||
|
||||
|
||||
|
||||
> দিমিত্রি সশনিকভের স্ক্রিনশট
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# সংক্ষিপ্ত বিবরণ
|
||||
|
||||
|
||||
|
||||
|
||||
> স্কেচনোট করেছেন [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
|
|||
|
|
@ -15,7 +15,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
CLIP-এর মূল ধারণা হলো টেক্সট প্রম্পটের সাথে একটি ইমেজ তুলনা করা এবং নির্ধারণ করা যে ইমেজটি প্রম্পটের সাথে কতটা সঙ্গতিপূর্ণ।
|
||||
|
||||
|
||||
|
||||
|
||||
> *ছবি [এই ব্লগ পোস্ট](https://openai.com/blog/clip/) থেকে নেওয়া হয়েছে*
|
||||
|
||||
|
|
@ -31,7 +31,7 @@ CLIP মডেল/লাইব্রেরি [OpenAI GitHub](https://github.com
|
|||
|
||||
ধরা যাক আমাদের ইমেজগুলোকে বিড়াল, কুকুর এবং মানুষের মধ্যে শ্রেণীবদ্ধ করতে হবে। এই ক্ষেত্রে, আমরা মডেলটিকে একটি ইমেজ এবং একটি সিরিজ টেক্সট প্রম্পট দিতে পারি: "*একটি বিড়ালের ছবি*", "*একটি কুকুরের ছবি*", "*একটি মানুষের ছবি*"। ফলাফল হিসেবে প্রাপ্ত ৩টি সম্ভাবনার ভেক্টরে আমরা সর্বোচ্চ মানের ইনডেক্সটি নির্বাচন করব।
|
||||
|
||||
|
||||
|
||||
|
||||
> *ছবি [এই ব্লগ পোস্ট](https://openai.com/blog/clip/) থেকে নেওয়া হয়েছে*
|
||||
|
||||
|
|
@ -55,13 +55,13 @@ VQGAN সম্পর্কে আরও জানতে [Taming Transformers](h
|
|||
|
||||
VQGAN এবং সাধারণ GAN-এর মধ্যে একটি গুরুত্বপূর্ণ পার্থক্য হলো, সাধারণ GAN যেকোনো ইনপুট ভেক্টর থেকে একটি ভালো ইমেজ তৈরি করতে পারে, কিন্তু VQGAN সম্ভবত একটি অসঙ্গতিপূর্ণ ইমেজ তৈরি করবে। তাই, ইমেজ তৈরির প্রক্রিয়াকে আরও নির্দেশনা দিতে হবে, এবং এটি CLIP ব্যবহার করে করা যেতে পারে।
|
||||
|
||||
|
||||
|
||||
|
||||
টেক্সট প্রম্পটের সাথে সঙ্গতিপূর্ণ একটি ইমেজ তৈরি করতে, আমরা কিছু র্যান্ডম এনকোডিং ভেক্টর দিয়ে শুরু করি যা VQGAN-এর মাধ্যমে একটি ইমেজ তৈরি করে। তারপর CLIP ব্যবহার করে একটি লস ফাংশন তৈরি করা হয় যা দেখায় ইমেজটি টেক্সট প্রম্পটের সাথে কতটা সঙ্গতিপূর্ণ। এরপর লক্ষ্য হলো এই লসকে কমানো, ব্যাক প্রোপাগেশন ব্যবহার করে ইনপুট ভেক্টর প্যারামিটারগুলো সামঞ্জস্য করা।
|
||||
|
||||
VQGAN+CLIP বাস্তবায়নের জন্য একটি চমৎকার লাইব্রেরি হলো [Pixray](http://github.com/pixray/pixray)
|
||||
|
||||
 |  | 
|
||||
 |  | 
|
||||
----|----|----
|
||||
প্রম্পট থেকে তৈরি ছবি *একটি বই সহ তরুণ পুরুষ সাহিত্য শিক্ষকের একটি ক্লোজআপ জলরঙের প্রতিকৃতি* | প্রম্পট থেকে তৈরি ছবি *একটি কম্পিউটার সহ তরুণ নারী কম্পিউটার বিজ্ঞান শিক্ষকের একটি ক্লোজআপ তেল প্রতিকৃতি* | প্রম্পট থেকে তৈরি ছবি *একটি ব্ল্যাকবোর্ডের সামনে বৃদ্ধ পুরুষ গণিত শিক্ষকের একটি ক্লোজআপ তেল প্রতিকৃতি*
|
||||
|
||||
|
|
|
|||
|
|
@ -1,8 +1,8 @@
|
|||
<!--
|
||||
CO_OP_TRANSLATOR_METADATA:
|
||||
{
|
||||
"original_hash": "4ed9993bca581850c983c95d5a3f57eb",
|
||||
"translation_date": "2025-12-24T23:47:07+00:00",
|
||||
"original_hash": "14816e97d79b296c87811724f7785923",
|
||||
"translation_date": "2026-01-01T11:27:30+00:00",
|
||||
"source_file": "README.md",
|
||||
"language_code": "br"
|
||||
}
|
||||
|
|
@ -19,116 +19,117 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
[](https://mybinder.org/v2/gh/microsoft/ai-for-beginners/HEAD)
|
||||
[](https://gitter.im/Microsoft/ai-for-beginners?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge)
|
||||
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
# Inteligência Artificial para Iniciantes - Um Currículo
|
||||
|
||||
||
|
||||
||
|
||||
|:---:|
|
||||
| AI For Beginners - _Sketchnote por [@girlie_mac](https://twitter.com/girlie_mac)_ |
|
||||
|
||||
Explore o mundo da **Inteligência Artificial** (IA) com nosso currículo de 12 semanas e 24 lições! Inclui lições práticas, questionários e laboratórios. O currículo é voltado para iniciantes e aborda ferramentas como TensorFlow e PyTorch, além de ética em IA
|
||||
Explore o mundo de **Inteligência Artificial** (IA) com nosso currículo de 12 semanas e 24 aulas! Inclui aulas práticas, questionários e laboratórios. O currículo é voltado para iniciantes e abrange ferramentas como TensorFlow e PyTorch, além de ética em IA
|
||||
|
||||
### 🌐 Suporte a Múltiplos Idiomas
|
||||
|
||||
### 🌐 Suporte Multilíngue
|
||||
|
||||
#### Suportado via GitHub Action (Automatizado e Sempre Atualizado)
|
||||
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE START -->
|
||||
[Árabe](../ar/README.md) | [Bengalês](../bn/README.md) | [Búlgaro](../bg/README.md) | [Birmanês (Myanmar)](../my/README.md) | [Chinês (Simplificado)](../zh/README.md) | [Chinês (Tradicional, Hong Kong)](../hk/README.md) | [Chinês (Tradicional, Macau)](../mo/README.md) | [Chinês (Tradicional, Taiwan)](../tw/README.md) | [Croata](../hr/README.md) | [Tcheco](../cs/README.md) | [Dinamarquês](../da/README.md) | [Holandês](../nl/README.md) | [Estoniano](../et/README.md) | [Finlandês](../fi/README.md) | [Francês](../fr/README.md) | [Alemão](../de/README.md) | [Grego](../el/README.md) | [Hebraico](../he/README.md) | [Híndi](../hi/README.md) | [Húngaro](../hu/README.md) | [Indonésio](../id/README.md) | [Italiano](../it/README.md) | [Japonês](../ja/README.md) | [Kannada](../kn/README.md) | [Coreano](../ko/README.md) | [Lituano](../lt/README.md) | [Malaio](../ms/README.md) | [Malayalam](../ml/README.md) | [Marathi](../mr/README.md) | [Nepalês](../ne/README.md) | [Pidgin Nigeriano](../pcm/README.md) | [Norueguês](../no/README.md) | [Persa (Farsi)](../fa/README.md) | [Polonês](../pl/README.md) | [Português (Brasil)](./README.md) | [Português (Portugal)](../pt/README.md) | [Punjabi (Gurmukhi)](../pa/README.md) | [Romeno](../ro/README.md) | [Russo](../ru/README.md) | [Sérvio (Cirílico)](../sr/README.md) | [Eslovaco](../sk/README.md) | [Esloveno](../sl/README.md) | [Espanhol](../es/README.md) | [Suaíli](../sw/README.md) | [Sueco](../sv/README.md) | [Tagalog (Filipino)](../tl/README.md) | [Tâmil](../ta/README.md) | [Telugo](../te/README.md) | [Tailandês](../th/README.md) | [Turco](../tr/README.md) | [Ucraniano](../uk/README.md) | [Urdu](../ur/README.md) | [Vietnamita](../vi/README.md)
|
||||
[Árabe](../ar/README.md) | [Bengalês](../bn/README.md) | [Búlgaro](../bg/README.md) | [Birmanês (Myanmar)](../my/README.md) | [Chinês (Simplificado)](../zh/README.md) | [Chinês (Tradicional, Hong Kong)](../hk/README.md) | [Chinês (Tradicional, Macau)](../mo/README.md) | [Chinês (Tradicional, Taiwan)](../tw/README.md) | [Croata](../hr/README.md) | [Tcheco](../cs/README.md) | [Dinamarquês](../da/README.md) | [Holandês](../nl/README.md) | [Estoniano](../et/README.md) | [Finlandês](../fi/README.md) | [Francês](../fr/README.md) | [Alemão](../de/README.md) | [Grego](../el/README.md) | [Hebraico](../he/README.md) | [Hindi](../hi/README.md) | [Húngaro](../hu/README.md) | [Indonésio](../id/README.md) | [Italiano](../it/README.md) | [Japonês](../ja/README.md) | [Kannada](../kn/README.md) | [Coreano](../ko/README.md) | [Lituano](../lt/README.md) | [Malaio](../ms/README.md) | [Malayalam](../ml/README.md) | [Marathi](../mr/README.md) | [Nepali](../ne/README.md) | [Pidgin Nigeriano](../pcm/README.md) | [Norueguês](../no/README.md) | [Persa (Farsi)](../fa/README.md) | [Polonês](../pl/README.md) | [Português (Brasil)](./README.md) | [Português (Portugal)](../pt/README.md) | [Punjabi (Gurmukhi)](../pa/README.md) | [Romeno](../ro/README.md) | [Russo](../ru/README.md) | [Sérvio (Cirílico)](../sr/README.md) | [Eslovaco](../sk/README.md) | [Esloveno](../sl/README.md) | [Espanhol](../es/README.md) | [Suaíli](../sw/README.md) | [Sueco](../sv/README.md) | [Tagalog (Filipino)](../tl/README.md) | [Tâmil](../ta/README.md) | [Telugu](../te/README.md) | [Tailandês](../th/README.md) | [Turco](../tr/README.md) | [Ucraniano](../uk/README.md) | [Urdu](../ur/README.md) | [Vietnamita](../vi/README.md)
|
||||
<!-- CO-OP TRANSLATOR LANGUAGES TABLE END -->
|
||||
|
||||
**Se você deseja que idiomas adicionais sejam suportados, veja a lista [aqui](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md)**
|
||||
**Se você deseja que idiomas adicionais sejam suportados, eles estão listados [aqui](https://github.com/Azure/co-op-translator/blob/main/getting_started/supported-languages.md)**
|
||||
|
||||
## Participe da Comunidade
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
## Junte-se à Comunidade
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
## O que você vai aprender
|
||||
## O que você aprenderá
|
||||
|
||||
**[Mapa Mental do Curso](http://soshnikov.com/courses/ai-for-beginners/mindmap.html)**
|
||||
|
||||
Neste currículo, você aprenderá:
|
||||
|
||||
* Diferentes abordagens para Inteligência Artificial, incluindo a antiga abordagem simbólica com **Representação do Conhecimento** e raciocínio ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence)).
|
||||
* **Redes Neurais** e **Aprendizado Profundo**, que estão no núcleo da IA moderna. Ilustraremos os conceitos por trás desses importantes tópicos usando código em duas das frameworks mais populares - [TensorFlow](http://Tensorflow.org) e [PyTorch](http://pytorch.org).
|
||||
* **Arquiteturas Neurais** para trabalhar com imagens e texto. Cobriremos modelos recentes, mas talvez estejamos um pouco defasados em relação ao estado da arte.
|
||||
* Diferentes abordagens de Inteligência Artificial, incluindo a abordagem simbólica "da velha guarda" com **Representação do Conhecimento** e raciocínio ([GOFAI](https://en.wikipedia.org/wiki/Symbolic_artificial_intelligence)).
|
||||
* **Redes Neurais** e **Deep Learning** (Aprendizado Profundo), que estão no núcleo da IA moderna. Vamos ilustrar os conceitos por trás desses tópicos importantes usando código em dois dos frameworks mais populares - [TensorFlow](http://Tensorflow.org) e [PyTorch](http://pytorch.org).
|
||||
* **Arquiteturas Neurais** para trabalhar com imagens e texto. Cobriremos modelos recentes, mas podemos estar um pouco defasados em relação ao estado da arte.
|
||||
* Abordagens de IA menos populares, como **Algoritmos Genéticos** e **Sistemas Multiagente**.
|
||||
|
||||
O que não cobriremos neste currículo:
|
||||
|
||||
> [Encontre todos os recursos adicionais para este curso em nossa coleção Microsoft Learn](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)
|
||||
|
||||
* Casos de uso de **IA nos Negócios**. Considere fazer a trilha de aprendizado [Introduction to AI for business users](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) no Microsoft Learn, ou [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum), desenvolvida em cooperação com [INSEAD](https://www.insead.edu/).
|
||||
* **Aprendizado de Máquina Clássico**, que é bem descrito em nosso [Machine Learning for Beginners Curriculum](http://github.com/Microsoft/ML-for-Beginners).
|
||||
* Aplicações práticas de IA construídas usando **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)**. Para isso, recomendamos que você comece com módulos do Microsoft Learn para [visão](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [processamento de linguagem natural](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Generative AI with Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** e outros.
|
||||
* Frameworks de ML em nuvem específicos, como [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum), ou [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum). Considere usar as trilhas de aprendizado [Build and operate machine learning solutions with Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) e [Build and Operate Machine Learning Solutions with Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum).
|
||||
* **IA Conversacional** e **Chat Bots**. Há uma trilha de aprendizado separada [Create conversational AI solutions](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum), e você também pode consultar [este post do blog](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/) para mais detalhes.
|
||||
* **Matemática Profunda** por trás do deep learning. Para isto, recomendamos [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) de Ian Goodfellow, Yoshua Bengio e Aaron Courville, que também está disponível online em [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/).
|
||||
* Casos de uso empresariais de **IA nos Negócios**. Considere fazer o caminho de aprendizado [Introdução à IA para usuários de negócios](https://docs.microsoft.com/learn/paths/introduction-ai-for-business-users/?WT.mc_id=academic-77998-bethanycheum) no Microsoft Learn, ou [AI Business School](https://www.microsoft.com/ai/ai-business-school/?WT.mc_id=academic-77998-bethanycheum), desenvolvido em cooperação com [INSEAD](https://www.insead.edu/).
|
||||
* **Machine Learning Clássico**, que é bem descrito em nosso [Currículo de Machine Learning para Iniciantes](http://github.com/Microsoft/ML-for-Beginners).
|
||||
* Aplicações práticas de IA construídas usando **[Cognitive Services](https://azure.microsoft.com/services/cognitive-services/?WT.mc_id=academic-77998-bethanycheum)**. Para isso, recomendamos que você comece pelos módulos do Microsoft Learn para [visão](https://docs.microsoft.com/learn/paths/create-computer-vision-solutions-azure-cognitive-services/?WT.mc_id=academic-77998-bethanycheum), [processamento de linguagem natural](https://docs.microsoft.com/learn/paths/explore-natural-language-processing/?WT.mc_id=academic-77998-bethanycheum), **[Generative AI with Azure OpenAI Service](https://learn.microsoft.com/en-us/training/paths/develop-ai-solutions-azure-openai/?WT.mc_id=academic-77998-bethanycheum)** e outros.
|
||||
* Frameworks de ML específicos de **Nuvem**, como [Azure Machine Learning](https://azure.microsoft.com/services/machine-learning/?WT.mc_id=academic-77998-bethanycheum), [Microsoft Fabric](https://learn.microsoft.com/en-us/training/paths/get-started-fabric/?WT.mc_id=academic-77998-bethanycheum), ou [Azure Databricks](https://docs.microsoft.com/learn/paths/data-engineer-azure-databricks?WT.mc_id=academic-77998-bethanycheum). Considere usar os caminhos de aprendizado [Build and operate machine learning solutions with Azure Machine Learning](https://docs.microsoft.com/learn/paths/build-ai-solutions-with-azure-ml-service/?WT.mc_id=academic-77998-bethanycheum) e [Build and Operate Machine Learning Solutions with Azure Databricks](https://docs.microsoft.com/learn/paths/build-operate-machine-learning-solutions-azure-databricks/?WT.mc_id=academic-77998-bethanycheum).
|
||||
* **IA Conversacional** e **Chat Bots**. Há um caminho de aprendizado separado [Create conversational AI solutions](https://docs.microsoft.com/learn/paths/create-conversational-ai-solutions/?WT.mc_id=academic-77998-bethanycheum), e você também pode consultar [este post do blog](https://soshnikov.com/azure/hello-bot-conversational-ai-on-microsoft-platform/) para mais detalhes.
|
||||
* **Matemática Avançada** por trás do deep learning. Para isso, recomendamos [Deep Learning](https://www.amazon.com/Deep-Learning-Adaptive-Computation-Machine/dp/0262035618) de Ian Goodfellow, Yoshua Bengio e Aaron Courville, que também está disponível online em [https://www.deeplearningbook.org/](https://www.deeplearningbook.org/).
|
||||
|
||||
Para uma introdução suave aos tópicos de _IA na Nuvem_ você pode considerar fazer a trilha de aprendizado [Get started with artificial intelligence on Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum).
|
||||
Para uma introdução suave aos tópicos de _IA na Nuvem_ você pode considerar fazer o caminho de aprendizado [Comece com inteligência artificial no Azure](https://docs.microsoft.com/learn/paths/get-started-with-artificial-intelligence-on-azure/?WT.mc_id=academic-77998-bethanycheum).
|
||||
|
||||
# Conteúdo
|
||||
|
||||
| | Link da Lição | PyTorch/Keras/TensorFlow | Laboratório |
|
||||
| | Link da Aula | PyTorch/Keras/TensorFlow | Laboratório |
|
||||
| :-: | :------------------------------------------------------------------------------------------------------------------------------------------: | :---------------------------------------------------------------------------------------------: | ------------------------------------------------------------------------------ |
|
||||
| 0 | [Configuração do Curso](./lessons/0-course-setup/setup.md) | [Configurar seu Ambiente de Desenvolvimento](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| 0 | [Configuração do Curso](./lessons/0-course-setup/setup.md) | [Configure seu Ambiente de Desenvolvimento](./lessons/0-course-setup/how-to-run.md) | |
|
||||
| I | [**Introdução à IA**](./lessons/1-Intro/README.md) | | |
|
||||
| 01 | [Introdução e História da IA](./lessons/1-Intro/README.md) | - | - |
|
||||
| II | **IA Simbólica** |
|
||||
| 02 | [Representação do Conhecimento e Sistemas Especialistas](./lessons/2-Symbolic/README.md) | [Sistemas Especialistas](./lessons/2-Symbolic/Animals.ipynb) / [Ontologia](./lessons/2-Symbolic/FamilyOntology.ipynb) /[Grafo de Conceitos](./lessons/2-Symbolic/MSConceptGraph.ipynb) | |
|
||||
| III | [**Introdução às Redes Neurais**](./lessons/3-NeuralNetworks/README.md) |||
|
||||
| 03 | [Perceptron](./lessons/3-NeuralNetworks/03-Perceptron/README.md) | [Notebook](./lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb) | [Lab](./lessons/3-NeuralNetworks/03-Perceptron/lab/README.md) |
|
||||
| 04 | [Multi-Layered Perceptron and Creating our own Framework](./lessons/3-NeuralNetworks/04-OwnFramework/README.md) | [Notebook](./lessons/3-NeuralNetworks/04-OwnFramework/OwnFramework.ipynb) | [Lab](./lessons/3-NeuralNetworks/04-OwnFramework/lab/README.md) |
|
||||
| 05 | [Intro to Frameworks (PyTorch/TensorFlow) and Overfitting](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Lab](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**Visão Computacional**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [Explore Computer Vision on Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 06 | [Introdução à Visão Computacional. OpenCV](./lessons/4-ComputerVision/06-IntroCV/README.md) | [Notebook](./lessons/4-ComputerVision/06-IntroCV/OpenCV.ipynb) | [Lab](./lessons/4-ComputerVision/06-IntroCV/lab/README.md) |
|
||||
| 07 | [Redes Neurais Convolucionais](./lessons/4-ComputerVision/07-ConvNets/README.md) & [Arquiteturas de CNN](./lessons/4-ComputerVision/07-ConvNets/CNN_Architectures.md) | [PyTorch](./lessons/4-ComputerVision/07-ConvNets/ConvNetsPyTorch.ipynb) /[TensorFlow](./lessons/4-ComputerVision/07-ConvNets/ConvNetsTF.ipynb) | [Lab](./lessons/4-ComputerVision/07-ConvNets/lab/README.md) |
|
||||
| 08 | [Redes pré-treinadas e Aprendizado por Transferência](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [Dicas de Treinamento](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Lab](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 03 | [Perceptron](./lessons/3-NeuralNetworks/03-Perceptron/README.md) | [Notebook](./lessons/3-NeuralNetworks/03-Perceptron/Perceptron.ipynb) | [Laboratório](./lessons/3-NeuralNetworks/03-Perceptron/lab/README.md) |
|
||||
| 04 | [Perceptron Multicamadas e Criando nosso próprio Framework](./lessons/3-NeuralNetworks/04-OwnFramework/README.md) | [Notebook](./lessons/3-NeuralNetworks/04-OwnFramework/OwnFramework.ipynb) | [Laboratório](./lessons/3-NeuralNetworks/04-OwnFramework/lab/README.md) |
|
||||
| 05 | [Introdução a Frameworks (PyTorch/TensorFlow) e Sobreajuste](./lessons/3-NeuralNetworks/05-Frameworks/README.md) | [PyTorch](./lessons/3-NeuralNetworks/05-Frameworks/IntroPyTorch.ipynb) / [Keras](./lessons/3-NeuralNetworks/05-Frameworks/IntroKeras.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Laboratório](./lessons/3-NeuralNetworks/05-Frameworks/lab/README.md) |
|
||||
| IV | [**Visão Computacional**](./lessons/4-ComputerVision/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-computer-vision-pytorch/?WT.mc_id=academic-77998-cacaste) / [TensorFlow](https://docs.microsoft.com/learn/modules/intro-computer-vision-TensorFlow/?WT.mc_id=academic-77998-cacaste)| [Explore Visão Computacional no Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) |
|
||||
| 06 | [Introdução à Visão Computacional. OpenCV](./lessons/4-ComputerVision/06-IntroCV/README.md) | [Notebook](./lessons/4-ComputerVision/06-IntroCV/OpenCV.ipynb) | [Laboratório](./lessons/4-ComputerVision/06-IntroCV/lab/README.md) |
|
||||
| 07 | [Redes Neurais Convolucionais](./lessons/4-ComputerVision/07-ConvNets/README.md) & [Arquiteturas de CNN](./lessons/4-ComputerVision/07-ConvNets/CNN_Architectures.md) | [PyTorch](./lessons/4-ComputerVision/07-ConvNets/ConvNetsPyTorch.ipynb) /[TensorFlow](./lessons/4-ComputerVision/07-ConvNets/ConvNetsTF.ipynb) | [Laboratório](./lessons/4-ComputerVision/07-ConvNets/lab/README.md) |
|
||||
| 08 | [Redes Pré-treinadas e Transferência de Aprendizado](./lessons/4-ComputerVision/08-TransferLearning/README.md) and [Dicas de Treinamento](./lessons/4-ComputerVision/08-TransferLearning/TrainingTricks.md) | [PyTorch](./lessons/4-ComputerVision/08-TransferLearning/TransferLearningPyTorch.ipynb) / [TensorFlow](./lessons/3-NeuralNetworks/05-Frameworks/IntroKerasTF.ipynb) | [Laboratório](./lessons/4-ComputerVision/08-TransferLearning/lab/README.md) |
|
||||
| 09 | [Autoencoders e VAEs](./lessons/4-ComputerVision/09-Autoencoders/README.md) | [PyTorch](./lessons/4-ComputerVision/09-Autoencoders/AutoEncodersPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/09-Autoencoders/AutoencodersTF.ipynb) | |
|
||||
| 10 | [Redes Adversariais Generativas & Transferência de Estilo Artístico](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 11 | [Detecção de Objetos](./lessons/4-ComputerVision/11-ObjectDetection/README.md) | [TensorFlow](./lessons/4-ComputerVision/11-ObjectDetection/ObjectDetection.ipynb) | [Lab](./lessons/4-ComputerVision/11-ObjectDetection/lab/README.md) |
|
||||
| 10 | [Redes Gerativas Adversariais & Transferência de Estilo Artístico](./lessons/4-ComputerVision/10-GANs/README.md) | [PyTorch](./lessons/4-ComputerVision/10-GANs/GANPyTorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/10-GANs/GANTF.ipynb) | |
|
||||
| 11 | [Detecção de Objetos](./lessons/4-ComputerVision/11-ObjectDetection/README.md) | [TensorFlow](./lessons/4-ComputerVision/11-ObjectDetection/ObjectDetection.ipynb) | [Laboratório](./lessons/4-ComputerVision/11-ObjectDetection/lab/README.md) |
|
||||
| 12 | [Segmentação Semântica. U-Net](./lessons/4-ComputerVision/12-Segmentation/README.md) | [PyTorch](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationPytorch.ipynb) / [TensorFlow](./lessons/4-ComputerVision/12-Segmentation/SemanticSegmentationTF.ipynb) | |
|
||||
| V | [**Processamento de Linguagem Natural**](./lessons/5-NLP/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-pytorch/?WT.mc_id=academic-77998-cacaste) /[TensorFlow](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-TensorFlow/?WT.mc_id=academic-77998-cacaste) | [Explore Natural Language Processing on Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)|
|
||||
| 13 | [Representação de Texto. BoW/TF-IDF](./lessons/5-NLP/13-TextRep/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationTF.ipynb) | |
|
||||
| 14 | [Embutimentos semânticos de palavras. Word2Vec and GloVe](./lessons/5-NLP/14-Embeddings/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsTF.ipynb) | |
|
||||
| 15 | [Modelagem de Linguagem. Treinando seus próprios embeddings](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [Lab](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| V | [**Processamento de Linguagem Natural**](./lessons/5-NLP/README.md) | [PyTorch](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-pytorch/?WT.mc_id=academic-77998-cacaste) /[TensorFlow](https://docs.microsoft.com/learn/modules/intro-natural-language-processing-TensorFlow/?WT.mc_id=academic-77998-cacaste) | [Explore Processamento de Linguagem Natural no Microsoft Azure](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum)|
|
||||
| 13 | [Representação de Texto. Bag-of-Words/TF-IDF](./lessons/5-NLP/13-TextRep/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/13-TextRep/TextRepresentationTF.ipynb) | |
|
||||
| 14 | [Embeddings semânticos de palavras. Word2Vec e GloVe](./lessons/5-NLP/14-Embeddings/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/14-Embeddings/EmbeddingsTF.ipynb) | |
|
||||
| 15 | [Modelagem de Linguagem. Treinando seus próprios embeddings](./lessons/5-NLP/15-LanguageModeling/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-PyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/15-LanguageModeling/CBoW-TF.ipynb) | [Laboratório](./lessons/5-NLP/15-LanguageModeling/lab/README.md) |
|
||||
| 16 | [Redes Neurais Recorrentes](./lessons/5-NLP/16-RNN/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNPyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/16-RNN/RNNTF.ipynb) | |
|
||||
| 17 | [Redes Recorrentes Generativas](./lessons/5-NLP/17-GenerativeNetworks/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativePyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativeTF.ipynb) | [Lab](./lessons/5-NLP/17-GenerativeNetworks/lab/README.md) |
|
||||
| 17 | [Redes Recorrentes Gerativas](./lessons/5-NLP/17-GenerativeNetworks/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativePyTorch.ipynb) / [TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/17-GenerativeNetworks/GenerativeTF.ipynb) | [Laboratório](./lessons/5-NLP/17-GenerativeNetworks/lab/README.md) |
|
||||
| 18 | [Transformers. BERT.](./lessons/5-NLP/18-Transformers/README.md) | [PyTorch](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersPyTorch.ipynb) /[TensorFlow](https://github.com/microsoft/AI-For-Beginners/blob/main/lessons/5-NLP/18-Transformers/TransformersTF.ipynb) | |
|
||||
| 19 | [Reconhecimento de Entidades Nomeadas](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [Lab](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [Grandes Modelos de Linguagem, Programação de Prompts e Tarefas Few-Shot](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| 19 | [Reconhecimento de Entidades Nomeadas](./lessons/5-NLP/19-NER/README.md) | [TensorFlow](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/19-NER/NER-TF.ipynb) | [Laboratório](./lessons/5-NLP/19-NER/lab/README.md) |
|
||||
| 20 | [Modelos de Linguagem Grandes, Programação de Prompts e Tarefas Few-Shot](./lessons/5-NLP/20-LangModels/README.md) | [PyTorch](https://microsoft.github.io/AI-For-Beginners/lessons/5-NLP/20-LangModels/GPT-PyTorch.ipynb) | |
|
||||
| VI | **Outras Técnicas de IA** || |
|
||||
| 21 | [Algoritmos Genéticos](./lessons/6-Other/21-GeneticAlgorithms/README.md) | [Notebook](./lessons/6-Other/21-GeneticAlgorithms/Genetic.ipynb) | |
|
||||
| 22 | [Aprendizado por Reforço Profundo](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [Lab](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 22 | [Aprendizado por Reforço Profundo](./lessons/6-Other/22-DeepRL/README.md) | [PyTorch](./lessons/6-Other/22-DeepRL/CartPole-RL-PyTorch.ipynb) /[TensorFlow](./lessons/6-Other/22-DeepRL/CartPole-RL-TF.ipynb) | [Laboratório](./lessons/6-Other/22-DeepRL/lab/README.md) |
|
||||
| 23 | [Sistemas Multiagente](./lessons/6-Other/23-MultiagentSystems/README.md) | | |
|
||||
| VII | **Ética em IA** | | |
|
||||
| 24 | [Ética em IA e IA Responsável](./lessons/7-Ethics/README.md) | [Microsoft Learn: Responsible AI Principles](https://docs.microsoft.com/learn/paths/responsible-ai-business-principles/?WT.mc_id=academic-77998-cacaste) | |
|
||||
| 24 | [Ética em IA e IA Responsável](./lessons/7-Ethics/README.md) | [Microsoft Learn: Princípios de IA Responsável](https://docs.microsoft.com/learn/paths/responsible-ai-business-principles/?WT.mc_id=academic-77998-cacaste) | |
|
||||
| IX | **Extras** | | |
|
||||
| 25 | [Redes Multimodais, CLIP and VQGAN](./lessons/X-Extras/X1-MultiModal/README.md) | [Notebook](./lessons/X-Extras/X1-MultiModal/Clip.ipynb) | |
|
||||
| 25 | [Redes Multimodais, CLIP e VQGAN](./lessons/X-Extras/X1-MultiModal/README.md) | [Notebook](./lessons/X-Extras/X1-MultiModal/Clip.ipynb) | |
|
||||
|
||||
## Cada lição contém
|
||||
## Cada aula contém
|
||||
|
||||
* Material de leitura prévia
|
||||
* Notebooks Jupyter executáveis, que frequentemente são específicos para o framework (**PyTorch** ou **TensorFlow**). O notebook executável também contém muito material teórico, então para entender o tópico você precisa passar por pelo menos uma versão do notebook (ou PyTorch ou TensorFlow).
|
||||
* **Labs** disponíveis para alguns tópicos, que oferecem a oportunidade de tentar aplicar o material que você aprendeu a um problema específico.
|
||||
* Algumas seções contêm links para módulos do [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) que cobrem tópicos relacionados.
|
||||
* Notebooks Jupyter executáveis, que frequentemente são específicos ao framework (**PyTorch** ou **TensorFlow**). O notebook executável também contém muito material teórico, então para entender o tópico você precisa passar por pelo menos uma versão do notebook (seja PyTorch ou TensorFlow).
|
||||
* **Laboratórios** disponíveis para alguns tópicos, que oferecem a oportunidade de tentar aplicar o material que você aprendeu a um problema específico.
|
||||
* Algumas seções contêm links para módulos do [**MS Learn**](https://learn.microsoft.com/en-us/collections/7w28iy2xrqzdj0?WT.mc_id=academic-77998-bethanycheum) que abordam tópicos relacionados.
|
||||
|
||||
## Como Começar
|
||||
## Primeiros Passos
|
||||
|
||||
### 🎯 Novo em IA? Comece aqui!
|
||||
### 🎯 Novo em IA? Comece Aqui!
|
||||
|
||||
Se você é completamente novo em IA e quer exemplos rápidos e práticos, confira nossos [**Exemplos para Iniciantes**](./examples/README.md)! Estes incluem:
|
||||
|
||||
- 🌟 **Hello AI World** - Seu primeiro programa de IA (reconhecimento de padrões)
|
||||
- 🧠 **Simple Neural Network** - Construir uma rede neural do zero
|
||||
- 🖼️ **Image Classifier** - Classificar imagens com comentários detalhados
|
||||
- 💬 **Análise de Sentimento de Texto** - Analisar texto positivo/negativo
|
||||
- 🧠 **Simple Neural Network** - Construa uma rede neural do zero
|
||||
- 🖼️ **Image Classifier** - Classifique imagens com comentários detalhados
|
||||
- 💬 **Análise de Sentimento de Texto** - Analise texto positivo/negativo
|
||||
|
||||
These examples are designed to help you understand AI concepts before diving into the full curriculum.
|
||||
|
||||
### 📚 Configuração do Currículo Completo
|
||||
|
||||
- We have created a [setup lesson](./lessons/0-course-setup/setup.md) to help you with setting up your development environment. - For Educators, we have created a [curricula setup lesson](./lessons/0-course-setup/for-teachers.md) for you too!
|
||||
- How to [Run the code in a VSCode or a Codepace](./lessons/0-course-setup/how-to-run.md)
|
||||
- Criamos uma [lição de configuração](./lessons/0-course-setup/setup.md) para ajudar você a configurar seu ambiente de desenvolvimento. - Para Educadores, também criamos uma [lição de configuração do currículo](./lessons/0-course-setup/for-teachers.md) para você!
|
||||
- Como [Executar o código no VSCode ou em um Codepace](./lessons/0-course-setup/how-to-run.md)
|
||||
|
||||
Follow these steps:
|
||||
|
||||
|
|
@ -144,21 +145,21 @@ Join our [official AI Discord server](https://aka.ms/genai-discord?WT.mc_id=acad
|
|||
|
||||
If you have product feedback or questions whilst building visit our [Azure AI Foundry Developer Forum](https://aka.ms/foundry/forum)
|
||||
|
||||
## Questionários
|
||||
## Quizzes
|
||||
|
||||
> **Uma nota sobre os questionários**: All quizzes are contained in the Quiz-app folder in etc\quiz-app, or [Online Aqui](https://ff-quizzes.netlify.app/) They are linked from within the lessons the quiz app can be run locally or deployed to Azure; follow the instruction in the `quiz-app` folder. They are gradually being localized.
|
||||
> **Uma nota sobre os questionários**: Todos os questionários estão contidos na pasta Quiz-app em etc\quiz-app, ou [Online Aqui](https://ff-quizzes.netlify.app/) Eles estão vinculados nas lições; o aplicativo de questionários pode ser executado localmente ou implantado no Azure; siga as instruções na pasta `quiz-app`. Eles estão sendo gradualmente localizados.
|
||||
|
||||
## Ajuda Solicitada
|
||||
## Precisamos de Ajuda
|
||||
|
||||
Do you have suggestions or found spelling or code errors? Raise an issue or create a pull request.
|
||||
Você tem sugestões ou encontrou erros de ortografia ou de código? Abra uma issue ou crie um pull request.
|
||||
|
||||
## Agradecimentos Especiais
|
||||
|
||||
* **✍️ Autor Principal:** [Dmitry Soshnikov](http://soshnikov.com), PhD
|
||||
* **🔥 Editora:** [Jen Looper](https://twitter.com/jenlooper), PhD
|
||||
* **🔥 Editor:** [Jen Looper](https://twitter.com/jenlooper), PhD
|
||||
* **🎨 Ilustradora de sketchnote:** [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
* **✅ Criadores dos Questionários:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 Colaboradores Principais:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
* **✅ Criadora de Questionários:** [Lateefah Bello](https://github.com/CinnamonXI), [MLSA](https://studentambassadors.microsoft.com/)
|
||||
* **🙏 Contribuidores Principais:** [Evgenii Pishchik](https://github.com/Pe4enIks)
|
||||
|
||||
## Outros Currículos
|
||||
|
||||
|
|
@ -187,7 +188,7 @@ Nossa equipe produz outros currículos! Confira:
|
|||
|
||||
---
|
||||
|
||||
### Aprendizado Principal
|
||||
### Aprendizado Fundamental
|
||||
[](https://aka.ms/ml-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/datascience-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
[](https://aka.ms/ai-beginners?WT.mc_id=academic-105485-koreyst)
|
||||
|
|
@ -204,19 +205,19 @@ Nossa equipe produz outros currículos! Confira:
|
|||
[](https://github.com/microsoft/CopilotAdventures?WT.mc_id=academic-105485-koreyst)
|
||||
<!-- CO-OP TRANSLATOR OTHER COURSES END -->
|
||||
|
||||
## Obtendo Ajuda
|
||||
## Obter Ajuda
|
||||
|
||||
If you get stuck or have any questions about building AI apps. Join fellow learners and experienced developers in discussions about MCP. It's a supportive community where questions are welcome and knowledge is shared freely.
|
||||
Se você ficar preso ou tiver quaisquer dúvidas sobre como construir aplicativos de IA, junte-se a outros aprendizes e desenvolvedores experientes em discussões sobre MCP. É uma comunidade de apoio onde perguntas são bem-vindas e o conhecimento é compartilhado livremente.
|
||||
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
[](https://discord.gg/nTYy5BXMWG)
|
||||
|
||||
If you have product feedback or errors while building visit:
|
||||
Se você tiver feedback do produto ou erros durante o desenvolvimento, visite:
|
||||
|
||||
[](https://aka.ms/foundry/forum)
|
||||
[](https://aka.ms/foundry/forum)
|
||||
|
||||
---
|
||||
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER START -->
|
||||
**Isenção de responsabilidade**:
|
||||
Este documento foi traduzido usando o serviço de tradução por IA [Co-op Translator](https://github.com/Azure/co-op-translator). Embora nos esforcemos para alcançar a precisão, esteja ciente de que traduções automatizadas podem conter erros ou imprecisões. O documento original em seu idioma nativo deve ser considerado a fonte autorizada. Para informações críticas, recomenda-se tradução profissional realizada por um humano. Não nos responsabilizamos por quaisquer mal-entendidos ou interpretações equivocadas decorrentes do uso desta tradução.
|
||||
Este documento foi traduzido usando o serviço de tradução por IA [Co-op Translator](https://github.com/Azure/co-op-translator). Embora nos esforcemos para garantir a precisão, esteja ciente de que traduções automatizadas podem conter erros ou imprecisões. O documento original em seu idioma nativo deve ser considerado a fonte autorizada. Para informações críticas, recomenda-se tradução profissional feita por um humano. Não nos responsabilizamos por quaisquer mal-entendidos ou interpretações incorretas decorrentes do uso desta tradução.
|
||||
<!-- CO-OP TRANSLATOR DISCLAIMER END -->
|
||||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Introdução à IA
|
||||
|
||||
|
||||
|
||||
|
||||
> Sketchnote por [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
@ -19,7 +19,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
Originalmente, os computadores foram inventados por [Charles Babbage](https://en.wikipedia.org/wiki/Charles_Babbage) para operar com números seguindo um procedimento bem definido - um algoritmo. Os computadores modernos, embora significativamente mais avançados que o modelo original proposto no século XIX, ainda seguem a mesma ideia de cálculos controlados. Assim, é possível programar um computador para fazer algo se soubermos a sequência exata de passos necessários para alcançar o objetivo.
|
||||
|
||||
|
||||
|
||||
|
||||
> Foto por [Vickie Soshnikova](http://twitter.com/vickievalerie)
|
||||
|
||||
|
|
@ -46,7 +46,7 @@ Para mais informações, consulte **[Inteligência Artificial Geral](https://en.
|
|||
|
||||
Um dos problemas ao lidar com o termo **[Inteligência](https://en.wikipedia.org/wiki/Intelligence)** é que não há uma definição clara para esse termo. Pode-se argumentar que inteligência está conectada ao **pensamento abstrato** ou à **autoconsciência**, mas não conseguimos defini-la adequadamente.
|
||||
|
||||
|
||||
|
||||
|
||||
> [Foto](https://unsplash.com/photos/75715CVEJhI) por [Amber Kipp](https://unsplash.com/@sadmax) do Unsplash
|
||||
|
||||
|
|
@ -98,13 +98,13 @@ Alternativamente, podemos tentar modelar os elementos mais simples dentro de nos
|
|||
|
||||
> | E o ML? | |
|
||||
> |--------------|-----------|
|
||||
> | Parte da Inteligência Artificial que se baseia no aprendizado do computador para resolver um problema com base em alguns dados é chamada de **Machine Learning**. Não consideraremos o aprendizado de máquina clássico neste curso - recomendamos o currículo separado [Machine Learning para Iniciantes](http://aka.ms/ml-beginners). |  |
|
||||
> | Parte da Inteligência Artificial que se baseia no aprendizado do computador para resolver um problema com base em alguns dados é chamada de **Machine Learning**. Não consideraremos o aprendizado de máquina clássico neste curso - recomendamos o currículo separado [Machine Learning para Iniciantes](http://aka.ms/ml-beginners). |  |
|
||||
|
||||
## Um Breve Histórico da IA
|
||||
|
||||
A Inteligência Artificial começou como um campo no meio do século XX. Inicialmente, o raciocínio simbólico era a abordagem predominante, e isso levou a uma série de sucessos importantes, como sistemas especialistas – programas de computador capazes de atuar como especialistas em alguns domínios de problemas limitados. No entanto, logo ficou claro que essa abordagem não escala bem. Extrair o conhecimento de um especialista, representá-lo em um computador e manter essa base de conhecimento precisa acaba sendo uma tarefa muito complexa e cara demais para ser prática em muitos casos. Isso levou ao chamado [Inverno da IA](https://en.wikipedia.org/wiki/AI_winter) na década de 1970.
|
||||
|
||||
<img alt="Breve Histórico da IA" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a0264357672b0884cf3a220fbafe35c65d70b2c3805f7bf5e.br.png" width="70%"/>
|
||||
<img alt="Breve Histórico da IA" src="../../../../translated_images/history-of-ai.7e83efa70b537f5a.br.png" width="70%"/>
|
||||
|
||||
> Imagem por [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
@ -124,7 +124,7 @@ Da mesma forma, podemos ver como a abordagem para criar “programas que falam
|
|||
* Assistentes modernos, como Cortana, Siri ou Google Assistant, são todos sistemas híbridos que usam redes neurais para converter fala em texto e reconhecer nossa intenção, e então empregam algum raciocínio ou algoritmos explícitos para realizar as ações necessárias.
|
||||
* No futuro, podemos esperar um modelo completamente baseado em redes neurais para lidar com diálogos por conta própria. As recentes redes neurais da família GPT e [Turing-NLG](https://www.microsoft.com/research/blog/turing-nlg-a-17-billion-parameter-language-model-by-microsoft) mostram grande sucesso nisso.
|
||||
|
||||
<img alt="a evolução do teste de Turing" src="../../../../translated_images/turing-test-evol.4184696701293ead6de6e6441a659c62f0b119b342456987f531005f43be0b6d.br.png" width="70%"/>
|
||||
<img alt="a evolução do teste de Turing" src="../../../../translated_images/turing-test-evol.4184696701293ead.br.png" width="70%"/>
|
||||
> Imagem de Dmitry Soshnikov, [foto](https://unsplash.com/photos/r8LmVbUKgns) por [Marina Abrosimova](https://unsplash.com/@abrosimova_marina_foto), Unsplash
|
||||
|
||||
## Pesquisas Recentes em IA
|
||||
|
|
|
|||
|
|
@ -12,7 +12,7 @@
|
|||
"\n",
|
||||
"Neste exemplo, vamos implementar um sistema simples baseado em conhecimento para determinar um animal com base em algumas características físicas. O sistema pode ser representado pela seguinte árvore AND-OR (esta é uma parte da árvore completa, podemos facilmente adicionar mais regras):\n",
|
||||
"\n",
|
||||
"\n"
|
||||
"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Representação de Conhecimento e Sistemas Especialistas
|
||||
|
||||
|
||||
|
||||
|
||||
> Sketchnote por [Tomomi Imura](https://twitter.com/girlie_mac)
|
||||
|
||||
|
|
@ -41,7 +41,7 @@ Na maioria das vezes, não definimos estritamente o conhecimento, mas o alinhamo
|
|||
|
||||
Assim, o problema da **representação de conhecimento** é encontrar uma maneira eficaz de representar o conhecimento dentro de um computador na forma de dados, para torná-lo automaticamente utilizável. Isso pode ser visto como um espectro:
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem por [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
@ -94,7 +94,7 @@ Sintaxe de Bloco | Indentação | | |
|
|||
|
||||
Um dos primeiros sucessos da IA simbólica foram os chamados **sistemas especialistas** - sistemas computacionais projetados para agir como especialistas em um domínio de problema limitado. Eles eram baseados em uma **base de conhecimento** extraída de um ou mais especialistas humanos e continham um **motor de inferência** que realizava algum raciocínio sobre ela.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
---------------------------------------------|------------------------------------------------
|
||||
Estrutura simplificada do sistema neural humano | Arquitetura de um sistema baseado em conhecimento
|
||||
|
||||
|
|
@ -106,7 +106,7 @@ Os sistemas especialistas são construídos como o sistema de raciocínio humano
|
|||
|
||||
Como exemplo, vamos considerar o seguinte sistema especialista para determinar um animal com base em suas características físicas:
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem por [Dmitry Soshnikov](http://soshnikov.com)
|
||||
|
||||
|
|
|
|||
|
|
@ -1256,7 +1256,7 @@
|
|||
"* Baixa perda de treinamento - o modelo consegue aproximar bem os dados de treinamento, pois tem poder expressivo suficiente.\n",
|
||||
"* A perda de validação pode ser muito maior do que a perda de treinamento e pode começar a aumentar durante o treinamento - isso ocorre porque o modelo \"memoriza\" os pontos de treinamento e perde a \"visão geral\".\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> Nesta imagem, `x` representa os dados de treinamento, e `o` os dados de validação. À esquerda - modelo linear (uma camada), que aproxima bem a natureza dos dados. À direita - modelo com overfitting, que aproxima perfeitamente os dados de treinamento, mas deixa de fazer sentido com qualquer outro dado (o erro de validação é muito alto).\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -58,7 +58,7 @@ Overfitting é um conceito extremamente importante em aprendizado de máquina, e
|
|||
|
||||
Considere o seguinte problema de aproximar 5 pontos (representados por `x` nos gráficos abaixo):
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-------------------------|--------------------------
|
||||
**Modelo linear, 2 parâmetros** | **Modelo não-linear, 7 parâmetros**
|
||||
Erro de treinamento = 5.3 | Erro de treinamento = 0
|
||||
|
|
@ -79,7 +79,7 @@ Erro de validação = 5.1 | Erro de validação = 20
|
|||
|
||||
Como você pode ver no gráfico acima, o overfitting pode ser detectado por um erro de treinamento muito baixo e um erro de validação alto. Normalmente, durante o treinamento, veremos tanto o erro de treinamento quanto o de validação começarem a diminuir, e então, em algum momento, o erro de validação pode parar de diminuir e começar a aumentar. Isso será um sinal de overfitting e um indicador de que provavelmente devemos parar o treinamento nesse ponto (ou pelo menos salvar um snapshot do modelo).
|
||||
|
||||
|
||||
|
||||
|
||||
## Como prevenir o overfitting
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Introdução às Redes Neurais
|
||||
|
||||
|
||||
|
||||
|
||||
Como discutimos na introdução, uma das formas de alcançar inteligência é treinar um **modelo computacional** ou um **cérebro artificial**. Desde meados do século XX, pesquisadores experimentaram diferentes modelos matemáticos, até que, nos últimos anos, essa abordagem se mostrou extremamente bem-sucedida. Esses modelos matemáticos do cérebro são chamados de **redes neurais**.
|
||||
|
||||
|
|
@ -36,13 +36,13 @@ Neste currículo, focaremos apenas em modelos de redes neurais.
|
|||
|
||||
Na biologia, sabemos que nosso cérebro é composto por células neurais (neurônios), cada uma delas tendo múltiplas "entradas" (dendritos) e uma única "saída" (axônio). Tanto os dendritos quanto os axônios podem conduzir sinais elétricos, e as conexões entre eles — conhecidas como sinapses — podem apresentar diferentes graus de condutividade, que são regulados por neurotransmissores.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
Neurônio Real *([Imagem](https://en.wikipedia.org/wiki/Synapse#/media/File:SynapseSchematic_lines.svg) da Wikipedia)* | Neurônio Artificial *(Imagem do Autor)*
|
||||
|
||||
Assim, o modelo matemático mais simples de um neurônio contém várias entradas X<sub>1</sub>, ..., X<sub>N</sub> e uma saída Y, e uma série de pesos W<sub>1</sub>, ..., W<sub>N</sub>. A saída é calculada como:
|
||||
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd767313e067719f400cec4b0e5090239c3e997c29f6789d4c3c263.br.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
<img src="../../../../translated_images/netout.1eb15eb76fd76731.br.png" alt="Y = f\left(\sum_{i=1}^N X_iW_i\right)" width="131" height="53" align="center"/>
|
||||
|
||||
onde f é alguma **função de ativação** não linear.
|
||||
|
||||
|
|
|
|||
|
|
@ -73,14 +73,14 @@ Em nosso [OpenCV Notebook](OpenCV.ipynb), damos alguns exemplos de quando a vis
|
|||
|
||||
* **Pré-processamento de uma fotografia de um livro em Braille**. Focamos em como podemos usar limiarização, detecção de características, transformação de perspectiva e manipulações NumPy para separar símbolos individuais em Braille para posterior classificação por uma rede neural.
|
||||
|
||||
 |  | 
|
||||
 |  | 
|
||||
----|-----|-----
|
||||
|
||||
> Imagem de [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
* **Detectando movimento em vídeo usando diferença de quadros**. Se a câmera estiver fixa, os quadros do feed da câmera devem ser bastante semelhantes entre si. Como os quadros são representados como arrays, apenas subtraindo esses arrays de dois quadros subsequentes obteremos a diferença de pixels, que deve ser baixa para quadros estáticos e se tornar maior quando houver movimento substancial na imagem.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem de [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
|
|
@ -89,7 +89,7 @@ Em nosso [OpenCV Notebook](OpenCV.ipynb), damos alguns exemplos de quando a vis
|
|||
- **Fluxo Óptico Denso** calcula o campo vetorial que mostra para cada pixel onde ele está se movendo.
|
||||
- **Fluxo Óptico Esparso** é baseado em pegar algumas características distintivas na imagem (por exemplo, bordas) e construir sua trajetória de quadro a quadro.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem de [OpenCV.ipynb](OpenCV.ipynb)
|
||||
|
||||
|
|
|
|||
|
|
@ -13,11 +13,11 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
|
||||
VGG-16 é uma rede que alcançou 92,7% de precisão na classificação top-5 do ImageNet em 2014. Ela possui a seguinte estrutura de camadas:
|
||||
|
||||
|
||||
|
||||
|
||||
Como você pode ver, a VGG segue uma arquitetura tradicional em forma de pirâmide, que é uma sequência de camadas de convolução e pooling.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem de [Researchgate](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493)
|
||||
|
||||
|
|
|
|||
|
|
@ -262,7 +262,7 @@
|
|||
"\n",
|
||||
"Assim, em uma CNN típica, haveria várias camadas de convolução, com camadas de pooling entre elas para diminuir as dimensões da imagem. Também aumentaríamos o número de filtros, porque à medida que os padrões se tornam mais avançados, há mais combinações interessantes possíveis que precisamos buscar.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Devido à diminuição das dimensões espaciais e ao aumento das dimensões de características/filtros, essa arquitetura também é chamada de **arquitetura piramidal**.\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -358,7 +358,7 @@
|
|||
"\n",
|
||||
"Assim, em uma CNN típica, haveria várias camadas de convolução, com camadas de pooling entre elas para diminuir as dimensões da imagem. Também aumentaríamos o número de filtros, porque, à medida que os padrões se tornam mais avançados, há mais combinações interessantes que precisamos procurar.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Devido à diminuição das dimensões espaciais e ao aumento das dimensões de características/filtros, essa arquitetura também é chamada de **arquitetura piramidal**.\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -17,7 +17,7 @@ Na vida real, queremos ser capazes de reconhecer objetos em uma imagem independe
|
|||
|
||||
Para extrair padrões, utilizaremos o conceito de **filtros convolucionais**. Como você sabe, uma imagem é representada por uma matriz 2D ou um tensor 3D com profundidade de cor. Aplicar um filtro significa que pegamos uma matriz relativamente pequena chamada **kernel do filtro**, e para cada pixel na imagem original calculamos a média ponderada com os pontos vizinhos. Podemos imaginar isso como uma pequena janela deslizando sobre toda a imagem e calculando a média de todos os pixels de acordo com os pesos na matriz do kernel do filtro.
|
||||
|
||||
 | 
|
||||
 | 
|
||||
----|----
|
||||
|
||||
> Imagem por Dmitry Soshnikov
|
||||
|
|
@ -38,7 +38,7 @@ O funcionamento das CNNs é baseado nas seguintes ideias importantes:
|
|||
* Podemos projetar a rede de forma que os filtros sejam treinados automaticamente
|
||||
* Podemos usar a mesma abordagem para encontrar padrões em características de alto nível, não apenas na imagem original. Assim, a extração de características pelas CNNs funciona em uma hierarquia de características, começando com combinações de pixels de baixo nível até combinações de alto nível de partes da imagem.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem de [um artigo de Hislop-Lynch](https://www.semanticscholar.org/paper/Computer-vision-based-pedestrian-trajectory-Hislop-Lynch/26e6f74853fc9bbb7487b06dc2cf095d36c9021d), baseado em [sua pesquisa](https://dl.acm.org/doi/abs/10.1145/1553374.1553453)
|
||||
|
||||
|
|
@ -55,9 +55,9 @@ A maioria das CNNs usadas para processamento de imagens segue a chamada arquitet
|
|||
|
||||
Como exemplo, vejamos a arquitetura do VGG-16, uma rede que alcançou 92,7% de precisão na classificação top-5 do ImageNet em 2014:
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem de [Researchgate](https://www.researchgate.net/figure/Vgg16-model-structure-To-get-the-VGG-NIN-model-we-replace-the-2-nd-4-th-6-th-7-th_fig2_335194493)
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@ Você precisa treinar uma rede neural convolucional para classificar diferentes
|
|||
|
||||
Usaremos o [Oxford-IIIT Pet Dataset](https://www.robots.ox.ac.uk/~vgg/data/pets/), que contém imagens de 37 diferentes raças de cães e gatos.
|
||||
|
||||
|
||||
|
||||
|
||||
Para baixar o conjunto de dados, use este trecho de código:
|
||||
|
||||
|
|
|
|||
|
|
@ -50,7 +50,7 @@
|
|||
"\n",
|
||||
"Para visualizar o gato ideal, começaremos com uma imagem de ruído aleatório e tentaremos usar a técnica de otimização por descida de gradiente para ajustar a imagem de forma que a rede reconheça um gato.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"Aqui está nossa imagem inicial:\n"
|
||||
]
|
||||
|
|
|
|||
|
|
@ -29,7 +29,7 @@ Tanto o Keras quanto o PyTorch possuem funções para carregar facilmente pesos
|
|||
|
||||
Aqui estão características extraídas de uma imagem de um gato pela rede VGG-16:
|
||||
|
||||
|
||||
|
||||
|
||||
## Conjunto de Dados de Gatos vs. Cachorros
|
||||
|
||||
|
|
@ -48,19 +48,19 @@ Uma rede neural pré-treinada contém diferentes padrões em seu *cérebro*, inc
|
|||
|
||||
Uma abordagem que podemos adotar é começar com uma imagem aleatória e, em seguida, tentar usar a técnica de **otimização por descida de gradiente** para ajustar essa imagem de forma que a rede comece a pensar que é um gato.
|
||||
|
||||
|
||||
|
||||
|
||||
No entanto, se fizermos isso, receberemos algo muito semelhante a um ruído aleatório. Isso ocorre porque *existem muitas maneiras de fazer a rede pensar que a imagem de entrada é um gato*, incluindo algumas que não fazem sentido visualmente. Embora essas imagens contenham muitos padrões típicos de um gato, não há nada que as restrinja a serem visualmente distintas.
|
||||
|
||||
Para melhorar o resultado, podemos adicionar outro termo à função de perda, chamado de **perda de variação**. É uma métrica que mostra quão semelhantes são os pixels vizinhos da imagem. Minimizar a perda de variação torna a imagem mais suave e elimina o ruído, revelando padrões mais visualmente atraentes. Aqui está um exemplo de tais imagens "ideais", classificadas como gato e zebra com alta probabilidade:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*Gato Ideal* | *Zebra Ideal*
|
||||
|
||||
Uma abordagem semelhante pode ser usada para realizar os chamados **ataques adversariais** em uma rede neural. Suponha que queremos enganar uma rede neural e fazer um cachorro parecer um gato. Se pegarmos a imagem de um cachorro, que é reconhecida pela rede como um cachorro, podemos ajustá-la um pouco usando otimização por descida de gradiente até que a rede comece a classificá-la como um gato:
|
||||
|
||||
 | 
|
||||
 | 
|
||||
-----|-----
|
||||
*Imagem original de um cachorro* | *Imagem de um cachorro classificada como gato*
|
||||
|
||||
|
|
|
|||
|
|
@ -21,7 +21,7 @@
|
|||
"\n",
|
||||
"Como estamos treinando o autoencoder para capturar o máximo de informações da imagem original possível para uma reconstrução precisa, a rede tenta encontrar a melhor **representação** das imagens de entrada para capturar o significado.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"> Imagem do [blog do Keras](https://blog.keras.io/building-autoencoders-in-keras.html)\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@
|
|||
"\n",
|
||||
"Como estamos treinando o autoencoder para capturar o máximo de informações possíveis da imagem original para uma reconstrução precisa, a rede tenta encontrar o melhor **embedding** das imagens de entrada para capturar o significado.\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"*Imagem do [blog do Keras](https://blog.keras.io/building-autoencoders-in-keras.html)*\n",
|
||||
"\n",
|
||||
|
|
|
|||
|
|
@ -19,7 +19,7 @@ No entanto, podemos querer usar dados brutos (não rotulados) para treinar extra
|
|||
|
||||
Como estamos treinando um autoencoder para capturar o máximo de informações da imagem original possível para uma reconstrução precisa, a rede tenta encontrar a melhor **representação** das imagens de entrada para capturar seu significado.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [blog do Keras](https://blog.keras.io/building-autoencoders-in-keras.html)
|
||||
|
||||
|
|
|
|||
|
|
@ -13,7 +13,7 @@ Os modelos de classificação de imagens que abordamos até agora tomavam uma im
|
|||
|
||||
## [Quiz pré-aula](https://ff-quizzes.netlify.app/en/ai/quiz/21)
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [site YOLO v2](https://pjreddie.com/darknet/yolov2/)
|
||||
|
||||
|
|
@ -25,7 +25,7 @@ Suponha que queremos encontrar um gato em uma imagem. Uma abordagem muito ingên
|
|||
2. Executar a classificação de imagem em cada bloco.
|
||||
3. Os blocos que resultarem em uma ativação suficientemente alta podem ser considerados como contendo o objeto em questão.
|
||||
|
||||
|
||||
|
||||
|
||||
> *Imagem do [Notebook de Exercícios](ObjectDetection-TF.ipynb)*
|
||||
|
||||
|
|
@ -42,7 +42,7 @@ Você pode encontrar os seguintes conjuntos de dados para essa tarefa:
|
|||
* [PASCAL VOC](http://host.robots.ox.ac.uk/pascal/VOC/) - 20 classes
|
||||
* [COCO](http://cocodataset.org/#home) - Objetos Comuns em Contexto. 80 classes, caixas delimitadoras e máscaras de segmentação
|
||||
|
||||
|
||||
|
||||
|
||||
## Métricas de Detecção de Objetos
|
||||
|
||||
|
|
@ -50,7 +50,7 @@ Você pode encontrar os seguintes conjuntos de dados para essa tarefa:
|
|||
|
||||
Enquanto na classificação de imagens é fácil medir o desempenho do algoritmo, na detecção de objetos precisamos medir tanto a correção da classe quanto a precisão da localização inferida da caixa delimitadora. Para este último, usamos a chamada **Interseção sobre União** (IoU), que mede o quão bem duas caixas (ou duas áreas arbitrárias) se sobrepõem.
|
||||
|
||||
|
||||
|
||||
|
||||
> *Figura 2 de [este excelente post sobre IoU](https://pyimagesearch.com/2016/11/07/intersection-over-union-iou-for-object-detection/)*
|
||||
|
||||
|
|
@ -98,11 +98,11 @@ Existem duas grandes classes de algoritmos de detecção de objetos:
|
|||
|
||||
[R-CNN](http://islab.ulsan.ac.kr/files/announcement/513/rcnn_pami.pdf) usa [Selective Search](http://www.huppelen.nl/publications/selectiveSearchDraft.pdf) para gerar uma estrutura hierárquica de regiões ROI, que são então passadas por extratores de características CNN e classificadores SVM para determinar a classe do objeto, e regressão linear para determinar as coordenadas da *caixa delimitadora*. [Artigo Oficial](https://arxiv.org/pdf/1506.01497v1.pdf)
|
||||
|
||||
|
||||
|
||||
|
||||
> *Imagem de van de Sande et al. ICCV’11*
|
||||
|
||||
|
||||
|
||||
|
||||
> *Imagens de [este blog](https://towardsdatascience.com/r-cnn-fast-r-cnn-faster-r-cnn-yolo-object-detection-algorithms-36d53571365e)*
|
||||
|
||||
|
|
@ -110,7 +110,7 @@ Existem duas grandes classes de algoritmos de detecção de objetos:
|
|||
|
||||
Essa abordagem é semelhante à R-CNN, mas as regiões são definidas após as camadas de convolução terem sido aplicadas.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [Artigo Oficial](https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Girshick_Fast_R-CNN_ICCV_2015_paper.pdf), [arXiv](https://arxiv.org/pdf/1504.08083.pdf), 2015
|
||||
|
||||
|
|
@ -118,7 +118,7 @@ Essa abordagem é semelhante à R-CNN, mas as regiões são definidas após as c
|
|||
|
||||
A ideia principal dessa abordagem é usar uma rede neural para prever ROIs - a chamada *Rede de Proposta de Região*. [Artigo](https://arxiv.org/pdf/1506.01497.pdf), 2016
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [Artigo Oficial](https://arxiv.org/pdf/1506.01497.pdf)
|
||||
|
||||
|
|
@ -130,7 +130,7 @@ Este algoritmo é ainda mais rápido que o Faster R-CNN. A ideia principal é a
|
|||
1. As características são processadas por **Position-Sensitive Score Map**. Cada objeto de $C$ classes é dividido em regiões $k\times k$, e treinamos para prever partes dos objetos.
|
||||
1. Para cada parte das regiões $k\times k$, todas as redes votam pelas classes de objetos, e a classe de objeto com o maior número de votos é selecionada.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [Artigo Oficial](https://arxiv.org/abs/1605.06409)
|
||||
|
||||
|
|
@ -141,7 +141,7 @@ YOLO é um algoritmo de uma única passagem em tempo real. A ideia principal é
|
|||
* A imagem é dividida em regiões $S\times S$.
|
||||
* Para cada região, **CNN** prevê $n$ objetos possíveis, coordenadas da *caixa delimitadora* e *confiança*=*probabilidade* * IoU.
|
||||
|
||||
|
||||
|
||||
|
||||
> Imagem do [Artigo Oficial](https://arxiv.org/abs/1506.02640)
|
||||
|
||||
|
|
|
|||
|
|
@ -9,7 +9,7 @@ CO_OP_TRANSLATOR_METADATA:
|
|||
-->
|
||||
# Visão Computacional
|
||||
|
||||
|
||||
|
||||
|
||||
Nesta seção, vamos aprender sobre:
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue