Events

CDS Seminar: Socially-aware ML via Influence Functions

Speaker: Ashia Wilson

Location: On-Line
Videoconference link: https://nyu.zoom.us/j/91634861736?pwd=TW4rWmxzdXJMRzRyZ1FHdGdRaytQZz09

Date: Wednesday, May 4, 2022

The influence function is commonly used in data science for a variety of tasks including (but not limited to) model explainability, fairness audits, robustness (model validation, outlier detection, and model selection) and privacy (data removal). In this talk, I will provide a broad overview of its use cases and then highlight some recent innovations which extend its usage to a wider class of machine learning models. I will also point to some fundamental limitations for its reliable use in several tasks and point to some important open problems.