Statistics 315a: Modern Statistical Learning

Lectures

• Time: Tuesday / Thursday, 1:30pm – 2:50pm

• Place: McCullough Building, Rm 115

Course resources

• Staff email list: stats315a-win2526-staff appropriate symbol lists.stanford.edu

• We will use Ed for discussion

• We will use Gradescope and Canvas for grading and assignment submission (assignments to provide requisite detail)

• Instructor contact: btrippe appropriate symbol stanford.edu

Instructor

Brian Trippe

• Office hour: Tuesday, 2:50pm-3:55pm in CoDa E232

Teaching assistants

Henry Smith

• Office hour: Friday 3:00pm-5:00pm in Sequoia Hall, Room 200

Ian Christopher Tanoh

• Office hour: Monday 1:00pm-2:00pm in CoDa B43

Course overview and learning goals

This course covers statistical techniques underlying modern machine learning. We hope to address several learning objectives.

• Build familiarity with common approaches in machine learning and their statistical foundations

• Recognize, given a dataset and analysis goal, appropriate ML tools to apply

• Practice applying and evaluating machine learning methods using Python and Pytorch

• Understand modern “large-scale” machine learning systems for text and image generation by building core components and simplified implementations

We will work towards these goals while covering several topics. Roughly two thirds of the course focuses explicitly on supervised learning. It covers:

• Prediction methods (linear models, trees and forests, deep neural networks)

• Model evaluation (cross-validation, calibration, conformal prediction)

The rest of the course covers statistical aspects of other topics in modern statistical learning:

• Optimization (convexity, stochastic methods, adaptive metrics)

• Generative models (variational autoencoders, diffusion models, and language models)

These topics have a central role in modern approaches to prediction problems.

Throughout the course, we will also discuss applications and touch on practical aspects of successful machine learning. These include: data-splitting and avoiding train/test contamination, outliers, and distribution shift. These aspects will appear in hands-on implementation assignments as well.

Prerequisites

Linear algebra, multivariate calculus, statistics (e.g., STATS 200 or 300A), and machine learning (CS/STATS 229 or equivalent), or permission of the instructor. Familiarity with Python and PyTorch is recommended.

What will we do in class?

Classes will be primarily lecture based, with material presented in class via slides and at the whiteboard (or drawn on an ipad). Some active learning activities will be incorporated, including follow-along programming tutorials (using python and pytorch) and “pair and share” pen and paper exercises.

Exams and review sessions

Grading

• Problem sets (20%),

• Peer grading (5%)

• Midterm (30%)

• Final (45%)

Part of the assessment is peer-grading classmates’ homework assignments. A reason for this is that reviewing the published solutions shortly after the due date, and critically assessing the work of others can help to learn the material. A subset of peer grading will be evaluated for accuracy and graded at the end of the quarter.

Strong class participation may result in some (small) extra credit at the end of the quarter when determining grades.

Assignment solutions and late policy

Assignment solutions will be released online at the time of each deadline on Canvas. This is so that solutions may be reviewed while the work is fresh in mind. This deadline to be set as late as possible without interfering with the subsequent assignments and exams. Due to the above, late assignments are not accepted. To accommodate unforeseen circumstances, the score of the lowest assignment will be dropped.

Accommodations

If you have an accommodation, please fill this form (link) and upload your current Stanford OAE letter to the form on the link ASAP to submit it to the teaching team and Academic Accommodations Coordinator. For questions for requests associated with accommodations contact Academic Accommodations Coordinator Zachary Rozman (zrozman@stanford.edu) or Teaching Assistant Ian Tanoh.

Midterm accommodations require notice of 10 calendar days (before February 1st). Exam accommodations require notice by Monday, 2 March 2026.

Technology policy

We strongly recommend taking notes (using a pen or pencil) in class, rather than typing them or not taking notes.

The first reason is for yourself. Our best education research suggests that this improves student learning substantially; for example, see the classic experimental study by Mueller and Oppenheimer, The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking. There has been followup work on this, including work with brain imaging.

A second reason is for others. Screens are distracting to others, particularly in the rare cases that one happens to switch to a tab containing something other than their notes. If you require a keyboard and screen in class, you must sit in the last row to avoid distracting others. Tablets that sit flat on the desk are fine.

Attribution

Aspects of this course (and syllabus) are adapted from John Duchi’s instantiation in Winter 2025.