Overview

This seminar course explores topics in distributed and federated machine learning, where statistical models are learnt across multiple nodes, sometimes without centralizing data. Students will engage with classic results and advanced research topics on distributed optimization, efficient ML, privacy-preserving ML, and other societal considerations regarding fairness and robustness. 

In the first few lectures, we will cover distributed machine learning focusing on optimization algorithms, systems constraints, and implementation techniques. For the majority of the course, we will discuss the problem of data sharing, focusing on the federated learning application, including problem formulation, algorithms, their trustworthiness considerations, real-world applications, and connections with other areas such as MLSys, signal processing, and data markets. 

Through lecture notes, paper reading, classroom discussions, and hands-on projects, students will learn background knowledge of related research topics, gain deeper understandings on the research landscape, critically analyze existing approaches, and identify open challenges.

Logistics

Course time: TR 09:30 am - 10:50 am CT

Location: RY 255

Course Staff

Instructor

Tian Li (email: litian@uchicago.edu, office: JCL 211, office hours: by appointment)

TA

Lalchand Pandia (email: lcpandia@uchicago.edu, office hours: Mondays 3pm-4pm, location JCL 207)

Canvas

We will use Canvas for course announcements and discussions. Reading assignments and projects will be submitted and graded through GradeScope. Rather than emailing questions to the teaching staff, please post your questions on Canvas Discussion. You are responsible for keeping up with announcements posted on Canvas by the instructor or TA, including clarifications, deadlines, and forms.

Prerequisites

Students are expected to have taken an introductory course in machine learning or its equivalent class. There is no strict pre-requisite. 

Grading

The final grade for the course will be based on the following weights:

• 40% participation (25% for paper presentations and 15% for other classroom participation)

• 30% paper reading assignments

• 30% project 

The project is an open-ended research project, and is required for students who take this course for letter grades but optional otherwise. When students take this course for P/F and choose to opt out of projects, the grading breakdown would be 60% participation and 40% paper reading assignments. Students are expected to read papers before the class and lead or participate in the discussion during the class. Each project can be done alone, or in groups of at most two with contributions of each member clearly identified in the final report. This course counts as a Data Science, ML and AI, or Mathematics and Statistics elective for PhD students. 

Course Policies

Paper Presentations

You are welcome to use any format (e.g., chalkboard, notes, or slides) or template. A required component is a list of at least three related discussion questions at the end of the presentation. 

Please see Canvas for detailed instructions and the signup link.

Paper Reading Assignments

Starting from week two, you are required to read the papers (usually two) in required reading of the schedule at least 15 minutes before the start of class when those papers will be discussed. Paper reviews should be submitted via Gradescope before the deadline (see late policies). Your paper reviews should consist of 3 sections per paper:

• What is the problem and why the problem is challenging?

• What is new in the proposed approach (i.e., what are the insights)?

• Connections with (all of or a subset of) other paper(s) (either in required reading or optional reading) listed in the same row of the schedule

Project

A course project is required for students who take this course for letter grades. The course project grading is as follows  (30 points in total):

• Project final presentation (20 points). Detailed breakdown is as below.

  • Introduction and overview: 4 points

  • Problem statement: 3 points

  • Related work: 3 points

  • Proposed system/method/theory: 4 points

  • Evaluation or discussions: 4 points

  • Q&A: 2 points

• Two-page project report (10 points). Detailed breakdown is scaled proportionally as in the final presentation above, but without Q&As.

  • Please specify the contribution of each team member (if more than one) clearly in a separate paragraph of the report. You can use any reasonable template for the final report.

Late Policies

Your lowest paper reading homework score will not be counted towards your final grade. This policy allows you to miss an assignment, but only one. This should be used to cover illness, family emergencies, job interviews, or any other extenuating circumstance, but no more than once. Late submissions will lose 10% of the available points per day late, and zero point after being 10 days late.

Pass/Fail Grading

A grade of P is given only for work of C quality or higher. You must request Pass/Fail grading prior to the day of the last lecture.

Use of AI tools

Use of AI tools such as ChatGPT are not allowed in for classroom participation or paper reading assignments. For the final project, we use the same policy as ICML 2025. Please email the intrstructor if you need further clarification.

Schedule