TRAK: Attributing Model Behavior at Scale
Authors
Sung Min Park, Kristian Georgiev, Andrew Ilyas, Guillaume Leclerc, Aleksander Madry
Venue
International Conference on Machine Learning (ICML)
Abstract
Introduces TRAK (Tracing with the Randomly-projected After Kernel), a data attribution method that is both effective and computationally tractable for large-scale models by leveraging random projections.
Tags