用于视频推理的潜在视觉缓存
Latent Visual Cache for Video Reasoning
AI总结：
研究视频推理中视觉锚定衰减问题，提出潜在视频缓存Latent-VC插入解码器，通过监督对比缓存对齐等训练，在多视频基准测试中表现优于基线，能保留视觉证据提升推理。
摘要：
视频推理需要大型多模态模型基于密集证据，但现有系统存在视觉锚定衰减。为此引入Latent-VC，一种插入解码器的循环潜在视觉缓存，通过监督对比缓存对齐等训练，在六个视频基准测试中优于强基线，能保留视觉证据提升视频推理。
英文摘要：
Video reasoning requires Large Multimodal Models (LMMs) to remain grounded in dense evidence, yet existing systems largely adopt "read-once, generate-many" paradigm, in which visual grounding weakens during generation. This phenomenon has been widely observed and is known as Visual Anchoring Decay. To fill this gap, we introduce Latent Video Cache (Latent-VC), a recurrent latent visual cache inserted into the decoder to preserve compact visual memories throughout reasoning. The cache is trained with supervised contrastive cache alignment and vision-grounded GRPO with a latent grounding reward, while maintaining strict train-inference alignment through native decoder hidden states. Built on Qwen3.5-9B, Latent-VC consistently outperforms strong CoT and SFT+GRPO baselines across six video benchmarks, with especially clear gains on grounding-intensive and long-video tasks. In addition, it also achieves higher accuracy with substantially shorter responses, suggesting that latent visual caching improves video reasoning by preserving visual evidence rather than relying on longer textual chains.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
