Influence functions (IFs) are a powerful tool for detecting anomalous examples in large scale datasets. However, they are unstable when applied to deep networks. In this paper, we provide an explanation for the instability of IFs and develop a solution to this problem. We show that IFs are unreliable when the two data points belong to two different classes. Our solution leverages class information to improve the stability of IFs. Extensive experiments show that our modification significantly improves the performance and stability of IFs while incurring no additional computational cost.
@article{arxiv.2305.01384,
title = {Class based Influence Functions for Error Detection},
author = {Thang Nguyen-Duc and Hoang Thanh-Tung and Quan Hung Tran and Dang Huu-Tien and Hieu Ngoc Nguyen and Anh T. V. Dau and Nghi D. Q. Bui},
journal= {arXiv preprint arXiv:2305.01384},
year = {2023}
}
Comments
Thang Nguyen-Duc, Hoang Thanh-Tung, and Quan Hung Tran are co-first authors of this paper. 12 pages, 12 figures. Accepted to ACL 2023