no code implementations • 18 Apr 2024 • Yinzhu Jin, Matthew B. Dwyer, P. Thomas Fletcher
Our method is based on the principle that if a model is dependent on a feature, then removal of that feature should significantly harm its performance.
no code implementations • 24 Jul 2023 • Yinzhu Jin, Jonathan C. Garneau, P. Thomas Fletcher
This paper introduces feature gradient flow, a new technique for interpreting deep learning models in terms of features that are understandable to humans.