Metraipolis › archive › dc_4eea12de14ef

SingularClip: Preventing Spectral Collapse to Maintain Plasticity in Continual and Reinforcement Learning — arXiv:2608.18319

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.18319
License (as archived)
CC0-1.0
Archived
2026-10-18T13:53:12.711Z
SHA-256 of the archived text
ed8f88cf92f6e9a303535f1cf837a6e466051c83c19cc448311cf9fbd4468322
Stable URL of this page
https://metraipolis.com/documento/dc_4eea12de14ef

How to cite

SingularClip: Preventing Spectral Collapse to Maintain Plasticity in Continual and Reinforcement Learning — arXiv:2608.18319 — arXiv (en), https://arxiv.org/abs/2608.18319 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4eea12de14ef (Metraipolis archive dc_4eea12de14ef, sha256 ed8f88cf92f6e9a303535f1cf837a6e466051c83c19cc448311cf9fbd4468322)
BibTeX
@misc{dc4eea12de14ef,
  title = {SingularClip: Preventing Spectral Collapse to Maintain Plasticity in Continual and Reinforcement Learning — arXiv:2608.18319},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.18319},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4eea12de14ef; sha256 ed8f88cf92f6e9a303535f1cf837a6e466051c83c19cc448311cf9fbd4468322},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.18319. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Neural networks trained on nonstationary tasks frequently lose the ability to fit new targets, a phenomenon referred to as loss of plasticity. We identify a novel source of plasticity loss due to the growing anisotropy of weight matrices' singular values during training, and analyze this phenomenon both empirically and theoretically. To mitigate this issue, we introduce SingularClip, a procedure that periodically clips the singular values of all weight matrices. We show that SingularClip performs strongly against baselines across a range of tasks in both continual supervised learning and deep reinforcement learning.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-20 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.18319 — 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 2026-10-18T13:50:58.900Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-08-20?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).