Course guide

The current 25-meeting course spine

This course is organized around the published lecture decks. The short calculus bridge is a distinct meeting, which is why the current sequence has 25 meetings rather than the older 24- and 26-lecture plans still preserved in the repository history.

How the course is built

Each module answers one question and hands a working idea to the next:

Module Meetings Core question
Foundations 1–4 Why these losses, and how do we differentiate a network?
Optimization & training 5–7 How do we train deep networks reliably and make them generalize?
Vision 8–11 How do locality, hierarchy, and dense prediction change the model?
Sequences & language 12–15 How do models represent and generate ordered data?
Transformers 16–17 How does attention become a scalable architecture?
Representation & multimodal learning 18–19 How can data supervise itself, and how do images meet text?
Generative modeling 20–23 How do latent-variable, adversarial, and diffusion models generate?
Systems & synthesis 24–25 How do we serve models efficiently, and what ideas survive the whole course?

The complete titles and canonical PDFs are on Lecture decks.

What to do each week

  • Deck first. It contains the conceptual thread, a worked mechanism, and two short MCQ checkpoints with immediate answer slides.
  • Notebook second. Use the companion only where calculation or implementation makes the idea clearer.
  • Assignment third. Assignments extend already-seen code and value an explanation of your own measurements over a leaderboard number.

Materials

  • Lecture decks — the canonical source for the current sequence.
  • Notebooks — runnable companions and downloadable .ipynb files.
  • Assignments — programming work, policy, and submission expectations.
  • Resources — a deliberately short list of books, videos, papers, and interactives.
  • Logistics & grading ↗ — schedule, grading, deadlines, and FAQ for the August 2026 offering.

Course norms

Bring questions to the course Slack workspace. For programming work, use AI tools honestly and document meaningful assistance; the assessed work is your ability to explain your own plots, decisions, and failures. The detailed policy lives on the Assignments page.