What to Expect for Content

• Covers material from all modules/homeworks: regression and trees, gradient descent/logistic regression/perceptron/basic neural nets, kernels/SVM/PCA, generative models (EM, VAE, GAN), CNN/RNN, word vectors/attention/transformers.
• Similar in style to the theory questions on the homeworks. Not pure math derivations or memorized formulas — questions target key algorithmic points: what a method does, why it works, what happens if you change an assumption or a piece of the setup.
• Expect some questions that give you a small modification to a familiar algorithm/setup and ask you to reason through the consequence.

How to Prepare

• Review your own homework solutions and the lecture notes — focus on why each method works and where it breaks, not on re-memorizing formulas.
• You don't need to redo derivations/proofs — those are secondary. Make sure you understand the key algorithm steps and math results themselves (what they say and why they matter).
• Write your own study notes as you go — you're allowed to bring them (see Exam Format).
• Practice explaining each algorithm out loud in plain language; if you can't, you don't know it yet.

Exam Format

• Designed to take 1-1.5 hours; you'll get 3 hours, in class, on paper, so you don't have to rush.
• Open book, open notes, open HWs — all must be printed or handwritten on paper (no devices with internet access).
• Mix of short conceptual questions, algorithm tracing, and "modify this setup, what changes" questions.

Grading

The final exam is about 15% of the final grade.

• Partial credit throughout — a partially correct, well-reasoned answer is worth more than a blank one.
• Emphasis is on demonstrating understanding of the key ideas, not on exact/perfect answers.