A Versatile Influence Function for Data Attribution with Non-Decomposable Loss

2024 article deng2024vif ○ Not yet verified

Authors

Junwei Deng, Weijing Tang, Jiaqi W. Ma

Venue

arXiv preprint

Abstract

Proposes Versatile Influence Function (VIF) designed to fully leverage auto-differentiation, eliminating case-specific derivations. Demonstrated across Cox regression for survival analysis, node embedding for network analysis, and listwise learning-to-rank, with estimates closely resembling leave-one-out retraining while being up to 10^3 times faster.

BibTeX

Local Entry

@article{deng2024vif,
  title = {A Versatile Influence Function for Data Attribution with Non-Decomposable Loss},
  author = {Junwei Deng and Weijing Tang and Jiaqi W. Ma},
  year = {2024},
  journal = {arXiv preprint},
  url = {https://arxiv.org/abs/2412.01335},
  abstract = {Proposes Versatile Influence Function (VIF) designed to fully leverage auto-differentiation, eliminating case-specific derivations. Demonstrated across Cox regression for survival analysis, node embedding for network analysis, and listwise learning-to-rank, with estimates closely resembling leave-one-out retraining while being up to 10^3 times faster.}
}

From AUTO:S2

@article{deng2024vif,
  title = {A Versatile Influence Function for Data Attribution with Non-Decomposable Loss},
  author = {Junwei Deng and Weijing Tang and Jiaqi W. Ma},
  year = {2024},
  journal = {International Conference on Machine Learning},
  doi = {10.48550/arXiv.2412.01335}
}