Metraipolis › archive › dc_46bbd3bd6c0d

Unlearning by Extrapolating a Shrink-and-Heal Lookahead — arXiv:2611.00374

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.00374
License (as archived): CC0-1.0
Archived: 2026-11-03T11:11:40.508Z
SHA-256 of the archived text: a0fd8d4749e217ec57c0cf45fd7859f47348ad18927e399fa2f409e5bdcecc1f
Stable URL of this page: https://metraipolis.com/documento/dc_46bbd3bd6c0d

How to cite

Unlearning by Extrapolating a Shrink-and-Heal Lookahead — arXiv:2611.00374 — arXiv (en), https://arxiv.org/abs/2611.00374 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_46bbd3bd6c0d (Metraipolis archive dc_46bbd3bd6c0d, sha256 a0fd8d4749e217ec57c0cf45fd7859f47348ad18927e399fa2f409e5bdcecc1f)

BibTeX

@misc{dc46bbd3bd6c0d,
  title = {Unlearning by Extrapolating a Shrink-and-Heal Lookahead — arXiv:2611.00374},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.00374},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_46bbd3bd6c0d; sha256 a0fd8d4749e217ec57c0cf45fd7859f47348ad18927e399fa2f409e5bdcecc1f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00374. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Unlearning aims to make a large language model behave like one retrained without its forget set. Prevailing likelihood-suppression objectives, however, lack an explicit forgetting target: they continually reward lower forget-set likelihood, making forgetting sensitive to the optimization budget. We address this limitation by estimating the direction toward the oracle without training it. Our key idea is to shrink the fine-tuned model toward its pre-fine-tuning weights and briefly re-learn the retain set, thereby removing the forget-specific contribution; in the linearized regime, this shrink-and-heal lookahead moves the model a fixed fraction of the way to the retrained model, its unique fixed point. Our method, SHIFT, extrapolates the lookahead's per-token changes into distillation targets and iteratively refines them. Evaluated against eleven baselines under a unified OpenUnlearning protocol, SHIFT achieves the highest aggregate score across all TOFU settings and the closest oracle alignment, 3.7x closer on average than the next-best method; on MUSE, it preserves the most retain knowledge. Under larger optimization budgets, baselines overshoot the oracle, while SHIFT remains near it, even under relearning attacks.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.00374 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-03T10:59:52.380Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).