( subject to changes: based on Tue & Thur classes)
Jan
17 - Course Overview and MLP Refresher
19 - MLP Lab, AI Agents and Motivation for Meta-learning
24 - MLP Agents Lab. Few-shot Learning (FSL) and Siamese Networks.
31 - Introduction to Meta-learning and Matching Networks
Feb
2 - Model-agnostic Meta-learning
HW1 - MAML Agent
7 - Introduction to Recurrent Neural Networks & Transformers
9 - Thinking like Transformers; RNNs and Transformers Lab
14 - Model-based Meta-learning and & In-context Learning
HW2 - CNP Agent
16 - Online and Continual Learning
21 - Tutorial Session for Q&A (by TA)
HW1 Due
23 - Introduction to Reinforcement Learning and RL-based AI Agents
28 (BITS only) - Reinforcement Learning Lab (Stable Baselines)
Mar
2 (BITS only) - Overview of Papers, Project Discussion, Scoping and Assignment
14 (IIITD & IITD only) - Reinforcement Learning Lab (Stable Baselines)
16 (IIITD & IITD only) - Overview of Papers, Project Discussion, Scoping and Assignment
21 - Meta Reinforcement Learning: Lab (Env for Projects) and RL^2
23- Many-shot Meta-RL: Algorithm Distillation
HW2 Due
28 - RL & Supervised Learning: [Meta-] Imitation-learning, Inverse-RL & Lab
30 - Overview of Language Models & Lab (From scratch, Hugging Face & OpenAI )
Apr
6 -: Tutorial for discussing projects (TAs)
11 - Meta-ICL, Fast-weight Language Models, TabLLM (Project options 1)
13 - Online Learning using Winnow/GLNs/FF Algorithm (Project option 2)
18 - Decision Transformer (Project Option 3), Hebbian-RL (Project option 4), Hebbian-augmented Training
20 - Introduction to Diffusion Models & Lab (Hugging Face)
25 - Planning using Diffusion Models (Project option 5)
27 - Wrap-up and Research Questions
May
2 - Projects Due
11 - Exam
Note: while some labs have been specifically called out, most lectures will also involve hands-on activities.
So students attending classes in-person are required to bring their laptops
Regarding projects: six options are listed above; in case a more than one student/group ends up doing the same project, they will each be given different data/tasks for which to use the same technique, and they may be evaluated in a competitive manner.