LLM Unlearning via Loss Adjustment with Only Forget Data

2024 inproceedings wang2024flat ○ Not yet verified

Authors

Yaxuan Wang, Jiaheng Wei, Chris Yuhao Liu, Jinlong Pang, Quan Liu, Ankit Parag Shah, Yujia Bao, Yang Liu, Wei Wei

Venue

ICLR 2025

Abstract

FLAT is a loss adjustment approach which maximizes f-divergence between the available template answer and the forget answer with respect to the forget data. Demonstrates superior unlearning performance compared to existing methods while minimizing impact on retained capabilities, tested on Harry Potter dataset and MUSE Benchmark.

BibTeX

Local Entry

@inproceedings{wang2024flat,
  title = {LLM Unlearning via Loss Adjustment with Only Forget Data},
  author = {Yaxuan Wang and Jiaheng Wei and Chris Yuhao Liu and Jinlong Pang and Quan Liu and Ankit Parag Shah and Yujia Bao and Yang Liu and Wei Wei},
  year = {2024},
  booktitle = {ICLR 2025},
  url = {https://arxiv.org/abs/2410.11143},
  abstract = {FLAT is a loss adjustment approach which maximizes f-divergence between the available template answer and the forget answer with respect to the forget data. Demonstrates superior unlearning performance compared to existing methods while minimizing impact on retained capabilities, tested on Harry Potter dataset and MUSE Benchmark.}
}

From AUTO:OPENALEX

@inproceedings{wang2024flat,
  title = {LLM Unlearning via Loss Adjustment with Only Forget Data},
  author = {Yaxuan Wang and Jiaheng Wei and Chris Yuhao Liu and Jinlong Pang and Quan Liu and Ankit Parag Shah and Yujia Bao and Yang Liu and Wei Wei},
  year = {2024},
  booktitle = {arXiv (Cornell University)},
  doi = {10.48550/arxiv.2410.11143}
}