Metraipolis › archive › dc_47d31cc3cff2
CoRe: Co-Evolving Reward Models for Mitigating Latent Reward Hacking in Video Diffusion Models — arXiv:2609.36245
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.36245
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T16:22:24.097Z
- SHA-256 of the archived text
9a0989e8ab266da127af0c69d31c9bb6dfc4be93beaa397efdee9219d85dd509- Stable URL of this page
https://metraipolis.com/documento/dc_47d31cc3cff2
How to cite
CoRe: Co-Evolving Reward Models for Mitigating Latent Reward Hacking in Video Diffusion Models — arXiv:2609.36245 — arXiv (en), https://arxiv.org/abs/2609.36245 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_47d31cc3cff2 (Metraipolis archive dc_47d31cc3cff2, sha256 9a0989e8ab266da127af0c69d31c9bb6dfc4be93beaa397efdee9219d85dd509)
BibTeX
@misc{dc47d31cc3cff2,
title = {CoRe: Co-Evolving Reward Models for Mitigating Latent Reward Hacking in Video Diffusion Models — arXiv:2609.36245},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.36245},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_47d31cc3cff2; sha256 9a0989e8ab266da127af0c69d31c9bb6dfc4be93beaa397efdee9219d85dd509},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36245. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Latent reward models (LRMs) enable efficient alignment of video diffusion models by scoring intermediate states directly in latent space. However, we find that optimizing against a fixed latent reward rapidly leads to latent reward hacking: the predicted reward stays high while perceptual and motion quality deteriorate. Our analysis identifies distributional escape as the central cause: within a few hundred updates, the generator moves beyond the reward model's training support, where its scores no longer reflect video quality. Based on this insight, we introduce CoRe, a co-evolving reward framework that treats latent-space alignment as a dynamic interaction between the generator and the reward model. Rather than optimizing against a stationary proxy, CoRe continually refits the reward model on the generator's current samples while anchoring it to real-video preferences, so the generator cannot gain reward by drifting away from the data. On Wan2.1-T2V-1.3B, experiments show that CoRe consistently improves generation quality over both the pretrained model and prior alignment methods, while avoiding the quality collapse of fixed-reward optimization. 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/2609.36245 — 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 2026-10-07T15:54:53.477Z 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).