DATA 443/643: Advanced Concepts in Large Language Models
This is a complementary course hub. Canvas is the official site for this semester. Questions: mousavi@american.edu.
The listed prerequisite is DATA 427/627. DATA 441/641 and DATA 442/642 are not required; overlap with those courses is intentional so this sequence is self-contained. The semester is project-based.
Syllabus
Class meetings
Two-hour class meetings. Canvas is official for submissions and any date changes. A typical meeting: teach the notes (worked examples at the board), play about 10–20 minutes of the listed video, then the discussion prompts. Labs fill remaining studio time or homework.
| Week | Topic | Notes |
|---|---|---|
| 1 | Neural networks and word embeddings | |
| 2 | Sequence models: RNNs, LSTMs, and GRUs | |
| 3 | Attention and transformers | |
| 4 | Multimodal foundations and vision transformers | |
| 5 | Vision–language models (CLIP and BLIP) | |
| 6 | Audio and cross-modal integration | |
| 7 | Scaling, efficiency, and deployment | |
| 8 | SFT, continual learning, and adapters | |
| 9 | RLHF and direct preference optimization | |
| 10 | Safety, editing, and retrieval-augmented training | |
| 11 | Generative models I: GANs | |
| 12 | Generative models II: diffusion | |
| 13 | Reasoning and chain-of-thought | |
| 14 | Tools, RAG, and final presentations | Final presentations |
| 15 | Final exam and remaining talks | Final exam |
Weekly materials
Open a week for its notes, labs, slides, videos, and discussion prompts. Canvas remains the official list of required reading.