Metraipolis › archive › dc_7bdaaaec3100

MIPS: Mitigating Mode Collapse in Reinforcement Learning for Large Outcome Spaces — arXiv:2610.19457

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.19457
License (as archived): CC0-1.0
Archived: 2026-10-15T14:15:35.875Z
SHA-256 of the archived text: b650c01e7c74cc4bc89ebca184ae0823ef4634f0ebe91baf914027a89478ce26
Stable URL of this page: https://metraipolis.com/documento/dc_7bdaaaec3100

How to cite

MIPS: Mitigating Mode Collapse in Reinforcement Learning for Large Outcome Spaces — arXiv:2610.19457 — arXiv (en), https://arxiv.org/abs/2610.19457 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7bdaaaec3100 (Metraipolis archive dc_7bdaaaec3100, sha256 b650c01e7c74cc4bc89ebca184ae0823ef4634f0ebe91baf914027a89478ce26)

BibTeX

@misc{dc7bdaaaec3100,
  title = {MIPS: Mitigating Mode Collapse in Reinforcement Learning for Large Outcome Spaces — arXiv:2610.19457},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19457},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7bdaaaec3100; sha256 b650c01e7c74cc4bc89ebca184ae0823ef4634f0ebe91baf914027a89478ce26},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19457. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Expected-reward maximization in reinforcement learning concentrates probability mass on a few outcomes, even in tasks that require many distinct, high-quality solutions, a failure known as mode collapse. Inverse Probability Scaling (IPS) counters this in group-based policy gradients such as GRPO by dividing each reward by the outcome's probability, estimated from its frequency within the sampled group. When the outcome space is large relative to the group, outcomes rarely repeat and these estimates become uninformative. Since an outcome can be constructed along many trajectories, its probability is a sum over all of them, and computing it directly is intractable. We propose Marginal IPS (MIPS), which performs marginal evaluation by combining the sampled trajectory with a learned model of the trajectories that lead to each outcome. On Hypergrids, where the exact sampling distribution is computable, MIPS stays near the reward-distribution target as the outcome space grows relative to the rollout group, while frequency-based IPS degrades. In molecular synthesis and phylogenetic inference, MIPS sustains diverse, high-quality sampling where GRPO and frequency-based IPS collapse, showing that MIPS serves as a reliable solution to mode collapse in RL.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.19457 — 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 2026-10-15T13:53:35.601Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).