|Talks|

Robust Deep Learning Under Distribution Shift

Visiting speaker
Past Talk
Zachary Chase Lipton
Assistant Professor, Carnegie Mellon University
Nov 7, 2019
10:00 am
Nov 7, 2019
10:00 am
In-person
4 Thomas More St
London E1W 1YW, UK
The Roux Institute
Room
100 Fore Street
Portland, ME 04101
Network Science Institute
2nd floor
Network Science Institute
11th floor
177 Huntington Ave
Boston, MA 02115
Network Science Institute
2nd floor
Room
58 St Katharine's Way
London E1W 1LP, UK

Talk recording

We might hope that when faced with unexpected inputs, well-designed software systems would fire off warnings. However, ML systems, which depend strongly on properties of their inputs (e.g. the i.i.d. assumption), tend to fail silently. Faced with distribution shift, we wish (i) to detect and (ii) to quantify the shift, and (iii) to correct our classifiers on the fly—when possible. This talk will describe a line of recent work on tackling distribution shift. First, I will focus on recent work on label shift, a classic problem, where strong assumptions enable principled methods. Then I will discuss how recent tools from generative adversarial networks have been appropriated (and misappropriated) to tackle dataset shift—characterizing and (partially) repairing a foundational flaw in the method.

About the speaker
Zachary Chase Lipton is an assistant professor at Carnegie Mellon University. His research spans both core machine learning methods and their social impact. This work addresses diverse application areas, including medical diagnosis, dialogue systems, and product recommendation. He is the founder of the Approximately Correct blog and an author of Dive Into Deep Learning, an interactive open-source book teaching deep learning through Jupyter notebooks. Find on Twitter (@zacharylipton) or GitHub (@zackchase).
Share this page:
Nov 07, 2019