Metraipolis › archive › dc_3584487ef045

Multi-Agent Self-Improving Reinforcement Learning for Video Reasoning — arXiv:2608.28675

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.28675
License (as archived): CC0-1.0
Archived: 2026-10-17T08:07:58.704Z
SHA-256 of the archived text: 09793570b4c8290390306c8a1b8ca23fe84b7e7a20bf8b6241f14320d1815eac
Stable URL of this page: https://metraipolis.com/documento/dc_3584487ef045

How to cite

Multi-Agent Self-Improving Reinforcement Learning for Video Reasoning — arXiv:2608.28675 — arXiv (en), https://arxiv.org/abs/2608.28675 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3584487ef045 (Metraipolis archive dc_3584487ef045, sha256 09793570b4c8290390306c8a1b8ca23fe84b7e7a20bf8b6241f14320d1815eac)

BibTeX

@misc{dc3584487ef045,
  title = {Multi-Agent Self-Improving Reinforcement Learning for Video Reasoning — arXiv:2608.28675},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.28675},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3584487ef045; sha256 09793570b4c8290390306c8a1b8ca23fe84b7e7a20bf8b6241f14320d1815eac},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.28675. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Video reasoning tasks such as grounded video question answering and temporal grounding require selecting temporal evidence that supports the query. In many current training setups, temporal supervision is applied through local objectives such as boundary regression or span generation, while verification is used mainly to rerank candidate segments at inference time. We study whether a frozen verifier can also guide training. Our multi-agent framework couples a trainable \emph{Grounder} with a frozen \emph{Verifier}: the Grounder samples candidate trajectories and evidence segments, the Verifier assigns query-conditioned segment scores, a group-relative policy-gradient objective favors trajectories that outperform their within-input peers, and a bootstrapped calibration loss steers temporal predictions toward verifier-preferred spans. Trained on source tasks and evaluated without target-dataset fine-tuning, a two-billion-parameter instantiation transfers zero-shot across grounded question answering, temporal grounding, and long-video question answering, reaching 28.7\% intersection-over-union and 25.4\% answer-grounding accuracy on a grounded-question-answering benchmark, 46.1\% intersection-over-union on a temporal-grounding benchmark, and 54.1\% on a long-video question-answering benchmark. Relative to a strong same-scale baseline, the gains are modest but consistent, with the clearest improvements on relevance-oriented metrics such as intersection-over-union and moderate-overlap recall. Within the tested benchmarks and transfer setting, the results support frozen verification as a training signal for evidence selection, while showing that strict boundary precision remains comparatively weaker. Code and models are available at this https URL

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.CL).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.28675 — 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 2026-10-17T07:47:52.897Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).