Metraipolis › archive › dc_85d4cb5ef3c0

STAR-GRPO: Canonical Anchoring and Reliability-First Advantages against Representation-Dependent Reward Hacking — arXiv:2609.36900

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36900
License (as archived)
CC0-1.0
Archived
2026-10-07T16:23:14.505Z
SHA-256 of the archived text
957227d9780436530b9ede44d4a49a7f20f066789d5cdc595708e4c47c45fdea
Stable URL of this page
https://metraipolis.com/documento/dc_85d4cb5ef3c0

How to cite

STAR-GRPO: Canonical Anchoring and Reliability-First Advantages against Representation-Dependent Reward Hacking — arXiv:2609.36900 — arXiv (en), https://arxiv.org/abs/2609.36900 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_85d4cb5ef3c0 (Metraipolis archive dc_85d4cb5ef3c0, sha256 957227d9780436530b9ede44d4a49a7f20f066789d5cdc595708e4c47c45fdea)
BibTeX
@misc{dc85d4cb5ef3c0,
  title = {STAR-GRPO: Canonical Anchoring and Reliability-First Advantages against Representation-Dependent Reward Hacking — arXiv:2609.36900},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36900},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_85d4cb5ef3c0; sha256 957227d9780436530b9ede44d4a49a7f20f066789d5cdc595708e4c47c45fdea},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36900. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reward hacking occurs when policy optimization exploits a brittle reward interface or an overly permissive proxy objective, improving the training score without improving the underlying response quality. This phenomenon is amplified in group-relative policy optimization: an unsupported reward can shift the group baseline and alter the updates of other rollouts, while post-hoc or purely relative weighting cannot represent group-wide uncertainty. We propose \emph{Self-Tuned Anchored Reliability Group-Relative Policy Optimization} (STAR-GRPO), a reliability-first advantage estimator based on paired assessments of the same rollout. STAR separates the quality signal from its learning influence: score disagreement determines rollout reliability, relative reliability enters a self-tuned robust location--scale fit before group normalization, and absolute group reliability attenuates the resulting bounded advantage. The analysis establishes coordinate and second-moment bounds, characterizes exact centering through the weighted location equation, and gives reliability-dependent attenuation guarantees for outlying rewards. We evaluate STAR-GRPO in two complementary reward-hacking regimes. In token-interface exploitation, STAR prevents runaway optimization of the deployed-interface score while improving the canonical quality signal. In rubric-proxy overoptimization for medical reasoning, STAR improves independent semantic evaluation, narrows the proxy--judge discrepancy, and reduces overclaim while optimizing the same task proxy. Together, these results show that reliability-first normalization offers a principled way to limit unsupported reward influence on both group baselines and policy updates, while retaining the task reward as the optimization target.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.36900 — 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 2026-10-07T15:54:53.562Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).