Metraipolis › archive › dc_2edf9b5cc4c9

Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training — arXiv:2609.15051

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.15051
License (as archived)
CC0-1.0
Archived
2026-10-15T22:53:34.272Z
SHA-256 of the archived text
7411413cab47d827a26227d73a18dc4f1cc7514c844383e527f7cebe0878ade2
Stable URL of this page
https://metraipolis.com/documento/dc_2edf9b5cc4c9

How to cite

Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training — arXiv:2609.15051 — arXiv (en), https://arxiv.org/abs/2609.15051 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2edf9b5cc4c9 (Metraipolis archive dc_2edf9b5cc4c9, sha256 7411413cab47d827a26227d73a18dc4f1cc7514c844383e527f7cebe0878ade2)
BibTeX
@misc{dc2edf9b5cc4c9,
  title = {Not All Prompts Are Equal: Exploration-Guided Prompt Scaffolding for Multimodal Reinforcement Post-Training — arXiv:2609.15051},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.15051},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2edf9b5cc4c9; sha256 7411413cab47d827a26227d73a18dc4f1cc7514c844383e527f7cebe0878ade2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.15051. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Training prompts in online reinforcement learning (RL) differ substantially in how informative they are for the current policy: some are already saturated while others are too difficult to yield reliable learning signals, yet both receive equal rollout budget under standard training. We propose an exploration-guided prompt scaffolding framework that adapts the training prompt distribution dynamically throughout RL post-training of multimodal large language models (MLLMs). Central to our approach is the $\textit{Exploration Potential Score} (EPS)$, a lightweight rollout-based proxy for prompt utility derived from KL-regularized policy improvement theory, computable directly from on-policy rollout statistics without additional overhead. Rather than discarding low-utility prompts, we use a teacher model to generate scaffolded rewrites that preserve the original task intent while making subsequent training more informative, reframing teacher supervision as training-data refinement rather than output imitation. Integrated with GRPO on Geo3K and MMK12, our method consistently outperforms the baseline on both in-domain and out-of-distribution benchmarks, achieving up to 9.7\% relative improvement in-domain and gains of 11.5\% on MathVision and 11.1\% on MMMU-Pro.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.15051 — 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 2026-10-15T22:29:03.790Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).