Metraipolis › archive › dc_68e7845fda86

Lightweight and Versatile Learned Optimization by Recombination of Gradient History — arXiv:2610.09604

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.09604
License (as archived)
CC0-1.0
Archived
2026-10-08T11:56:46.341Z
SHA-256 of the archived text
d19f994c940ac234812201f4671f493ae60ad720d9be536736b81367ee407411
Stable URL of this page
https://metraipolis.com/documento/dc_68e7845fda86

How to cite

Lightweight and Versatile Learned Optimization by Recombination of Gradient History — arXiv:2610.09604 — arXiv (en), https://arxiv.org/abs/2610.09604 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_68e7845fda86 (Metraipolis archive dc_68e7845fda86, sha256 d19f994c940ac234812201f4671f493ae60ad720d9be536736b81367ee407411)
BibTeX
@misc{dc68e7845fda86,
  title = {Lightweight and Versatile Learned Optimization by Recombination of Gradient History — arXiv:2610.09604},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.09604},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_68e7845fda86; sha256 d19f994c940ac234812201f4671f493ae60ad720d9be536736b81367ee407411},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.09604. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

This paper presents a lightweight and versatile learned optimizer that dynamically recombines gradient history, represented as averages over disjoint time spans. The optimizer reduces the prediction space to one scalar coefficient per gradient average, shared by multiple parameters. Progressively averaging older gradients minimizes memory cost of long history, while keeping their contributions independently accessible. A 37k-parameter network trained in 0.87 GPU-hours generalizes zero-shot to unseen tasks, lowering validation loss by 9.1% and 0.4% on BERT-Tiny and GPT-Tiny, and improving test accuracy over Adam by 3.5 %p on a Vision Transformer and by 2.7 %p on average across nine graph models, with FLOPs overhead as low as 0.3%.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.09604 — 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 2026-10-08T10:50:37.699Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).