This is an introductory deep learning course for computer science undergraduate students that comprehensively covers the mathematical foundations, architectural principles, and cutting-edge methodologies of deep learning—the core pillar of modern AI. Beginning with the fundamentals of machine learning, the curriculum explores core deep learning mechanisms in depth. Students will derive the backpropagation and gradient descent algorithms that form the bedrock of deep neural networks, and progress from CNNs and RNNs to the Transformer architectures powering today's state-of-the-art AI.

Breaking away from rigid theoretical instruction, this course will adopt a generative AI-based teaching and learning workflow to adapt flexibly to rapid technological paradigm shifts. Students will engage with frontline topics in industry and academia from diverse perspectives. Students will also implement and optimize real-world code using AI coding agents, and engage in real-time debates with AI on open-ended technical topics, equipping them with both theoretical depth and practical insight.

Students who successfully complete this course will achieve the following competencies:
- Mastery of Theoretical & Mathematical Principles: Clear understanding of the underlying mathematics in deep learning; ability to mathematically derive and explain forward and backward propagation in neural networks.
- Analysis of Modern Architectures & Trends: Ability to analyze core mechanisms of modern architectures (e.g., Transformers) and critically compare large-scale model scaling trends and advancing training algorithms from an engineering perspective.
- Advanced AI-Collaborative Implementation: Proficiency in designing core deep learning layers and training pipelines using "Vibe Coding" workflows (seamlessly integrating PyTorch with AI coding agents); ability to independently debug and optimize system bugs and performance bottlenecks.
- Critical Thinking & Technical Communication: Ability to logically critique the limitations of deep learning through discussions with peers and virtual AI panels; capability to clearly communicate complex technical concepts using visual and interactive elements.

While there are no formal prerequisites for this course, the following background knowledge is highly recommended:

- Foundational Mathematics: A basic grasp of high-school or introductory college-level calculus, linear algebra, probability theory, and statistics.
- Basic Programming Skills: Familiarity with Python syntax and basic programming logic.
- Basic Machine Learning Literacy: Familiarity with core concepts such as linear/logistic regression, training models, supervised learning, and regularization.
- AI Agent Prompting & Debugging Literacy: Experience of writing clear, structured instructions to guide an AI's behavior and being able to identify, troubleshoot, and fix common errors or unexpected outputs.

1. Introduction
2. Supervised learning
3. Shallow neural networks
4. Deep neural networks
5. Loss functions
6. Fitting models
7. Gradients and initialization
8. Measuring performance
9. Regularization
10. Convolutional networks
11. Residual networks
12. Transformers
13. Graph neural networks
14. Unsupervised learning
15. Generative adversarial networks
16. Normalizing flows
17. Variational autoencoders
18. Diffusion models
19. Reinforcement learning
20. Why does deep learning work?
21. Deep learning and ethics

- Standard Lectures: Foundational concepts and core principles will be delivered through standard academic lectures.
- Discussion Sessions: To foster critical thinking, each lecture will include a dedicated discussion session where students can engage with the topics and exchange ideas.
- Midterm and Final Examinations: There will be two comprehensive exams during the semester to assess the understanding of fundamental theories.
- Hands-on Experiments and Practical Skills using AI: The course will also offer hands-on practice sessions utilizing coding agents, and students will have to create a blog post and interactive playground, gaining hands-on experience in AI collaboration.

- The curriculum is intended for undergraduates, while we also permit the participation of graduate students who wish to build their foundations in deep learning.
- Thus priority will be given to undergraduate students in case of over-enrollment.
- The discussion sessions will help you build a deep intuition through peer-to-peer exchange.