Friday, September 15th, 2023
Logistics
- Date: Friday, September 15th
- In-person Location: Northwestern University: Levy Mayer Hall, LM104, Chicago.
- Register: Click here to register
Aravindan Vijayaraghavan, Northwestern University
Tentative Schedule
| 9:30AM CT | Breakfast |
| 9:45AM CT | Introduction to the Fall Special Program by Aravindan Vijayaraghavan |
| 10AM CT | Avrim Blum (TTIC) on Robustness of different flavors (an ML Theory / TCS perspective) |
| 10:30AM CT | Dan Linna (Northwestern) on CS+Law: Designing Trustworthy and Reliable Systems in Law and CS |
| 11AM CT | Cong Ma (U Chicago) on An introduction (and invitation) to learning under distribution shift |
| 11:30AM-11:50AM CT | Break |
| 11:50AM-12:30PM CT | Lightning talks (Saba Ahmadi, Jinshuo Dong, Gyorgy Turan, Aravindan Vijayaraghavan, Ren Wang, and Ming Zhong) |
| 12:30PM CT | Lunch |
| 1:45PM-3:45PM | Open problems session + brainstorming |
Titles and Abstracts
Avrim Blum, Toyota Technological Institute at Chicago
Speaker: Avrim Blum
Title: Robustness of different flavors (an ML Theory / TCS perspective)
Abstract: In this talk I will give a brief introduction to robustness of a few different flavors from a Machine Learning Theory and Theoretical Computer Science perspective. In general, the goal is to achieve good performance even in the presence of different forms of data corruption or model misspecification. This is typically approached by considering an adversary (real or imagined) with different abilities to corrupt data and different goals for doing so, and aiming to understand what kinds of guarantees the learner (or adversary) can achieve.
Daniel W. Linna Jr., Northwestern University
Speaker: Daniel W. Linna Jr.
Title: CS+Law: Designing Trustworthy and Reliable Systems in Law and CS
Abstract: TBA
Cong Ma, University of Chicago
Speaker: Cong Ma
Title: An introduction (and invitation) to learning under distribution shift
Abstract: Learning under distribution shift is concerned with the scenario when the training and test distribution for a model are mismatched. In this talk, I will provide a brief introduction to several lines of research in this area: domain adaptation, structured distribution shifts (e.g., covariate shift), and distributionally robust optimization. I will also discuss their pros and cons. In the end, I will talk about some of the challenges in the space of learning under distribution shift.