AI learning
动手学深度学习
Dive into Deep Learning (D2L) 2.0 is an open-source deep learning textbook designed specifically for Chinese readers.
What is 动手学深度学习?
Dive into Deep Learning (D2L) 2.0 is an open-source deep learning textbook designed specifically for Chinese readers.It serves not just as a theoretical guide but as a hands-on platform with complete code implementations, supporting major frameworks including PyTorch, NumPy/MXNet, TensorFlow, and PaddlePaddle.Adopted by over 500 universities in more than 70 countries, it aims to master core deep learning concepts through practical application.
Every section of the book provides executable Jupyter notebooks, allowing readers to freely modify code and hyperparameters for immediate feedback, thereby building real-world deep learning experience.The curriculum covers everything from linear regression and multilayer perceptrons to Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Transformers, and modern architectures like BERT, combining formulas, diagrams, and real datasets for interactive teaching.
In addition to the online documentation, printed editions (both black-and-white paperback and full-color hardcover) are available on platforms like JD.com and Dangdang.The project boasts an active community offering courseware, assignments, and teaching videos based on the PyTorch version (available on Bilibili and YouTube), facilitating a deep understanding of the entire workflow from data preprocessing to model deployment.
Key features
- Multi-FrameworkOfficial code implementations are provided for PyTorch, NumPy/MXNet, TensorFlow, and PaddlePaddle, allowing for comparative learning.
- Interactive JupyEvery section is an executable code unit, supporting local execution, Amazon SageMaker Studio Lab, SageMaker, or Google Colab.
- From Scratch & CKey algorithms provide both from-scratch low-level implementations to reveal principles and concise implementations using high-level APIs for engineering efficiency.
- Comprehensive CoCovers prerequisites, linear neural networks, CNNs, RNNs, attention mechanisms, Transformers, optimization algorithms, and applications in computer vision and NLP.
- Real-World KagglIncludes practical cases from real Kaggle competitions such as house price prediction, image classification (CIFAR-10), and dog breed recognition.
- Active CommunityBacked by over 200 community contributors, offering Chinese courseware, teaching videos, and forum support to help resolve learning challenges.
How to use it
- Visit the official website at zh.d2l.ai and select your preferred deep learning framework version (e.g., PyTorch or PaddlePaddle).
- Read the preface and installation guide to set up your local Python environment, or use the provided cloud environments like Google Colab.
- Study chapters in order, starting with prerequisites (linear algebra, calculus) and progressing to linear neural networks and multilayer perceptrons.
- Download and run the Jupyter notebooks corresponding to each chapter, trying to modify hyperparameters in the code to observe changes in model performance.
- Participate in community discussions linked at the end of each chapter, or watch accompanying teaching videos on Bilibili/YouTube to deepen understanding.
- Complete the practical projects in the book (such as Kaggle competition cases) to apply what you have learned to real-world datasets.
Where it fits
- University DeepServes as a textbook for undergraduate or graduate students, providing standardized courseware, assignments, and experimental code to lower teaching barriers.
- Engineer Skill EHelps developers with programming backgrounds deeply understand the underlying principles of deep learning, moving from calling APIs to understanding model architectures.
- Kaggle CompetitiLearn data preprocessing, feature engineering, and model tuning techniques through practical cases in the book to improve competition rankings.
- Academic ResearcProvides clear implementations and mathematical derivations of classic models (like ResNet, BERT) for researchers to facilitate reproduction and improvement.
Pricing and free quota
Free to read online; printed books are sold on platforms like JD.com and Dangdang.Specific pricing follows the official site.
Who it is for
- Deep learning beginners
- Computer science students
- AI algorithm engineers
- Data scientists
- University instructors
FAQ
What programming background do I need to study this book?
Basic Python programming skills are recommended.Chapter 2 'Prerequisites' reviews necessary linear algebra, calculus, and probability knowledge, but familiarity with Python and data manipulation libraries like NumPy is helpful.
Which deep learning framework version should I choose?
The PyTorch version is currently very popular and actively updated in the community, making it suitable for most beginners.If your company or school primarily uses TensorFlow or PaddlePaddle, you can choose those versions as the core concepts are universal.
How do I run the code in the book?
You can click the 'Jupyter Notebook' link at the top of each section to download the code and run it locally, or use cloud environments like Google Colab or Amazon SageMaker for one-click execution without local setup.
Is this book suitable for complete beginners?
It is suitable for readers with some programming background but unfamiliar with deep learning theory.Starting from simple linear regression and progressing gradually with many diagrams and code explanations, it is ideal for getting started.