Metraipolis › archive › dc_a265fcbd3228

Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry — arXiv:2610.20350

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.20350
License (as archived)
CC0-1.0
Archived
2026-10-15T14:23:50.217Z
SHA-256 of the archived text
0228aa98342e2c2945a2114a4112ca5a72a4811edc7e3c42667f1690c3a39d88
Stable URL of this page
https://metraipolis.com/documento/dc_a265fcbd3228

How to cite

Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry — arXiv:2610.20350 — arXiv (en), https://arxiv.org/abs/2610.20350 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a265fcbd3228 (Metraipolis archive dc_a265fcbd3228, sha256 0228aa98342e2c2945a2114a4112ca5a72a4811edc7e3c42667f1690c3a39d88)
BibTeX
@misc{dca265fcbd3228,
  title = {Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry — arXiv:2610.20350},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20350},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a265fcbd3228; sha256 0228aa98342e2c2945a2114a4112ca5a72a4811edc7e3c42667f1690c3a39d88},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20350. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Preference optimization ranks final answers and trajectory RL maximizes aggregate returns, yet neither credits the local change that turns a flawed reasoning state into a better one. We introduce Lifted State Policy Optimization (LSPO), which establishes this finite, cost-adjusted transition as the primary unit of learning. Each textual answer is augmented with a learned continuous coordinate, forming a lifted state that an energy function evaluates and orders. A shaped reward drives energy descent net of edit and step costs, and an internalization loss absorbs selected low-energy answers into the direct generator. On Qwen3.5-27B, the resulting single-pass policy reaches 71.8% aggregate accuracy across mathematics, science, code, and logic benchmarks, matching test-time reflection at 71.9% with 28% lower latency and 25.8% fewer tokens. The pattern holds across dense and mixture-of-experts backbones, and matched controls confirm that transition credit and the lifted coordinate each contribute, as does answer internalization. Our code is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.20350 — 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 2026-10-15T13:53:35.960Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).