Overview

This seminar course will cover recent advances in the theoretical foundations of (trustworthy) machine learning, including but not limited to topics like privacy, fairness, robustness, generalization, safety, and incentives. The course will consistent primarily of student-led presentations of classic and recent research papers, and will be shaped by students' interests.

Time & Location

MTh 11:35 – 1:15pm in 177 Huntington 6th Floor Conference Room

Note 1: This is not the official room assigned by the registrar!
Note 2: 177 Huntington has its own security system, please contact me to gain access

Instructor

Course Content

The course will cover a variety of themes around the theoretical foundations of (trustworthy) machine learning, as well as how those ideas are being applied or might be applied in practice. Each theme will involve a few meetings and I will solicit preferences between each theme. We may cover themes I haven't even anticipated yet, but the following (soon to be growing) list represents topics I am interested in that we may cover:

• Privacy for Statistics and ML
• Privacy attacks on ML systems
• Differential privacy
• Contextual integrity
• Private ML in practice
• Privacy for the Census
• Fairness and Decision-Making
• Individual and group fairness definitions
• Calibration and decision-making
• Causal notions of fairness
• Tradeoffs between fairnes notions
• Fair ML in practice
• Generalization and Statistical Validity
• Generalization theory
• Generalization in deep learning
• Generalization in adaptive data analysis
• The reproducibility crisis
• Robustness and Attacks on ML Systems
• Robust statistical inference
• Data poisoning
• Out-of-distribution generalization
• Data Deletion and Machine Unlearning
• Attempts at definitions
• Common methods and pitfalls
• Copyright
• Attempts at definitions
• Approaches to protecting copyright

Student Led Meetings

Each student will be responsible for leading one or more meetings (depending on the number of students we have). Before leading a meeting, students are encouraged to discuss the paper(s) they will be presenting informally with me and make a plan for how to lead the meeting effectively.

Grading

This is a PhD seminar course with the goal of exposing students to frontier research topics, and to build the skills of reading, discussing, and presenting research papers. As such, there is no set material you are supposed to come away from the course mastering, and there is no systematic evaluation process. All that is expected is to attends regularly, participate, and put effort into the presentations.

Resources

Since each topic will involve reading a number of classic and recent research papers (possibly with some relevant background material), I will compile a list of relevant papers and other resources here as we go.