Designing and Understanding Deep Neural Networks
Foundation models (FMs), including GPT-4, DALL·E 3, and Sora, have revolutionized several domains such as computer vision, language technology, and robotics. These deep learning (DL) models, trained on extensive and diverse datasets, are versatile across numerous downstream tasks, underpinning cutting-edge systems with their remarkable generative and few-shot learning capabilities. This comprehensive course offers students foundational knowledge in modeling and system design of FMs, supplemented by practical experience. The curriculum incorporates covers design principles, best practices, and optimization strategies. The course concludes with a four-week project, enabling students to devise a DL-based research project or application, addressing an issue of their interest.
Calendar
Basics
- Feb 17
- Lecture 1
Introduction- HW 1 out
- Feb 24
- Lecture 2
ML Review - Mar 3
- Lecture 3
Neural Networks - Mar 10
- Lecture 4
Optimization
Models
- Mar 17
- Lecture 5
Building Blocks- HW 2 out
- Mar 24
- Lecture 6
ConvNets- HW 1 due
- Mar 31
- Lecture 7
RNNs - Apr 7
- Lecture 8
Transformers- Project Proposal due
Applications
- Apr 14
- Lecture 9
Sequence to Sequence Models- HW 3 out
- Apr 21
- Lecture 10
Distribution Shift- HW 2 due
- Apr 28
- Lecture 11
Robustness - May 5
- Lecture 12
Generative Models - May 12
- Lecture 13
Self-Supervised Learning- HW 4 out
- May 19
- Lecture 14
Foundation Models- HW 3 due
Projects
- May 26
- Lecture 15
Project Presentation - June 2
- Lecture 16
Project Presentation - June 16
- Project Report due
- HW 4 due