| title | RepSelect: Robust LLM Unlearning via Representation Selectivity |
|---|---|
| authors | Filip Sondej, Yushi Yang, Adam Mahdi |
| date | 2026-06-15 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2606.17168 |
| affiliation | University of Oxford |
| upvotes | 4 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | Making large language models (LLMs) deeply forget specific knowledge and values without sacrificing general capabilities remains a central challenge in unlearning. However, current methods are easily reversed by fine-tuning or few-shot prompting, suggesting their forgetting is only shallow. We identify the root cause. Existing methods target representations shared with both the retain set and the subspace recovered by a fine-tuning attacker, making unlearning both disruptive to general capabilities and easy to reverse. We propose RepSelect (Representation Selectivity), isolates forget-set-specific representations by collapsing top principal components of weight gradients before each update, leaving general capabilities intact while limiting what fine-tuning can recover. We evaluate across two forget categories, biohazardous knowledge and abusive tendencies, and four model families spanning dense and Mixture-of-Experts architectures (Llama 3, Qwen 3.5, Gemma 4 E4B, DeepSeek V2 Lite). Compared to five popular baselines (GradDiff, NPO, SimNPO, RMU, UNDIAL), RepSelect achieves a 4-50x larger reduction in post-relearning answer accuracy than the strongest baseline, and is near-perfectly robust to few-shot prompting attacks. Targeting selective representations is thus an important step towards deep and robust LLM forgetting. |
cite
@misc{sondej2026repselect,
title={RepSelect: Robust LLM Unlearning via Representation Selectivity},
author={Filip Sondej and Yushi Yang and Adam Mahdi},
year={2026},
eprint={2606.17168},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2606.17168}
}