B9145: Topics in Trustworthy AI
Course Schedule
Date | Topics | Lecture Notes | Readings |
Jan 30 | Course Overview | Part 1: Course Overview, Part 2: Data-centric View of AI | CLIP |
Feb 6 | Language for Distribution Shifts | Slides, DRO | DISDE |
Feb 13 | Domain Adaptation | Underspecification, Tatsu Hashimoto's slides | A theory of learning from different domains |
Feb 20 | Invariance | Invariance, IRM, Domain Generalization | In Search of Lost Domain Generalization |
Feb 27 | LLM pre-training, SFT, RLHF | Overview, Finetuning, RLHF | GPT3 |
Mar 6 | LLM reasoning, inference-time operations | Slides | |
Mar 27 | Pre-training data and scaling laws | Pre-training Data, Scaling Laws | Chinchilla |
April 3 | Uncertainty quantification | Slides | List of references |
April 10 | Adaptive data collection: bandits | Slides | |
April 17 | Adaptive data collection: Bayesian optimization & active learning | BayesOpt, Active Learning | Bayesian RL |
April 24 | Adaptive data collection: MDP view & UQ as generative modeling, LLM moderation | Part 1: Bayesian Adaptive MDP, Part 2: Autoregressive generation as posterior inference, Part 3: Moderation | Uncertainty as missing data |
|
Email the instructor if you want access to lecture recordings.
|