Metraipolis › archive › dc_c6fcff573680

Neuralyzing the Trace: Selective Representation-Level Unlearning with Contrastive Sparse Autoencoders — arXiv:2609.31056

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.31056
License (as archived)
CC0-1.0
Archived
2026-10-08T01:38:31.404Z
SHA-256 of the archived text
bab95fe009779583b87b18fdea1af9d08ec9d22818099e470bd82ea11b8faab0
Stable URL of this page
https://metraipolis.com/documento/dc_c6fcff573680

How to cite

Neuralyzing the Trace: Selective Representation-Level Unlearning with Contrastive Sparse Autoencoders — arXiv:2609.31056 — arXiv (en), https://arxiv.org/abs/2609.31056 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_c6fcff573680 (Metraipolis archive dc_c6fcff573680, sha256 bab95fe009779583b87b18fdea1af9d08ec9d22818099e470bd82ea11b8faab0)
BibTeX
@misc{dcc6fcff573680,
  title = {Neuralyzing the Trace: Selective Representation-Level Unlearning with Contrastive Sparse Autoencoders — arXiv:2609.31056},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.31056},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_c6fcff573680; sha256 bab95fe009779583b87b18fdea1af9d08ec9d22818099e470bd82ea11b8faab0},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.31056. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Machine unlearning aims to remove targeted information while preserving a model's other abilities. In realistic settings, such as privacy requests under the EU GDPR, the target may be narrow, for example information associated with a single person. Behavioral forgetting alone may be insufficient, motivating interventions directly on internal representations. However, standard mechanistic-interpretability extractors are poorly selective for such targets. We identify an energy bias in reconstruction-based extraction, which favors dominant background structure over low-energy target-specific components. We introduce SCALPEL, a contrastive sparse autoencoder designed to learn more selective forget features. We show theoretically that contrastive training promotes target-selective features and that our selection score controls expected background knowledge perturbation. We validate SCALPEL experimentally on TOFU across Qwen, Llama, and Gemma, where it substantially improves over NMF and standard SAE interventions and is competitive with Gradient Difference and RMU, bridging mechanistic interpretability and fine-grained unlearning.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.31056 — 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 2026-10-08T01:12:14.331Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).