CSCI 567: Machine Learning (Fall 2026)

Course Description and Objectives

Machine learning powers everything from spam filters to large language models. This course covers the core techniques of supervised and unsupervised learning - how they work, why they work, and when they fail. We build up from linear models to deep networks and transformers, with a running focus on generalization: why does a model that fits its training data work (or not work) on data it has never seen? Some of the questions we will answer:

• how do we know a model actually works, and why do LLM benchmarks overstate progress (generalization and evaluation)
• how learning becomes optimization: loss functions and (stochastic) gradient descent
• how linear models fit nonlinear patterns (feature maps and kernel methods)
• why can heavily overparameterized models still generalize (double descent, implicit regularization)
• how neural networks are trained (backpropagation, Adam) and what makes it hard
• what is inside ChatGPT: self-supervised learning, transformers, scaling laws, and RLHF
• how to probe whether any of it is safe (alignment, interpretability)

Objectives: The course introduces standard statistical machine learning techniques for supervised and unsupervised learning, emphasizing conceptual understanding of the methods, their applications, and hands-on experience. Lectures focus on the fundamental ideas; the weekly discussion section mixes problem solving with hands-on Python/Colab labs.

Prerequisites

Undergraduate-level linear algebra, multi-variable calculus, probability and statistics, analysis of algorithms, and Python programming proficiency. The first discussion sections include a linear algebra and probability refresher.

Grading

1. Homeworks (30%): Four homeworks with written and programming components, done in pairs, submitted via Gradescope and Brightspace. You have 3 late days total across the semester, at most 1 per assignment. Each homework is due at 11:59pm on the Monday of the week whose discussion section reviews it (HW1 Sep 14, HW2 Sep 28, HW3 Nov 2, HW4 Nov 16), so late-day submissions still land before solutions are discussed.
2. Exam 1 (25%): In class on Oct 16.
3. Exam 2 (25%): Friday, Dec 11, 2:00-4:00pm (university-scheduled finals slot).
4. Course Project (20%): Done in groups of 4. Read the final project guide.
   • Project topic and team due: Oct 9
   • Project check-in (in discussion section): Oct 30
   • Project report review (in discussion section): Nov 20
   • Final report due: end of finals period (~Dec 17)

Active class contribution earns a discretionary bump.

Letter grades: there is no fixed point scale. Final letter grades are assigned at the end of the semester based on the overall score distribution, with guaranteed minimums: an overall score of 90 or above guarantees at least an A-, and 80 or above guarantees at least a B-. The curve can only help you - cutoffs may move down from these floors, but never up.

Resources

Each homework has two submissions: the PDF write-up, with answers to every question (for programming parts, include the printed numbers or screenshots, the plots, and your explanations), goes to Gradescope, where it is graded; a zip of your code goes to Brightspace.

There is no required textbook. Lecture slides, notes, and Colab notebooks will be posted in the schedule below as the semester progresses. Optional references:

• Kevin Murphy. Probabilistic Machine Learning: An Introduction. MIT Press, 2022. [link]
• Trevor Hastie, Robert Tibshirani, Jerome Friedman. The Elements of Statistical Learning. Springer, 2009. [link]
• Moritz Hardt, Benjamin Recht. Patterns, Predictions, and Actions. Princeton University Press, 2022. [link]
• Shai Shalev-Shwartz, Shai Ben-David. Understanding Machine Learning. Cambridge University Press, 2014. [link]

With permission from their instructors, parts of this course adapt material from Vatsal Sharan's CSCI 567 (Spring 2026, USC) and EPFL's CS-433 Machine Learning.

Course Schedule