# Watching Itself Watch: Self-Auditing Visual Reliance for Video Reasoning

DevFeed: [Watching Itself Watch: Self-Auditing Visual Reliance for Video Reasoning](<https://devfeed.tech/articles/watching-itself-watch-self-auditing-visual-reliance-for-video-reasoning-125483.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11949>)

Author: Rohan Dugad, Mi Luo, Kristen Grauman

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [multimodal-ai](<https://devfeed.tech/topics/multimodal-ai.md>), [grpo](<https://devfeed.tech/topics/grpo.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>)

Tags: [auditing](<https://devfeed.tech/tags/auditing.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [learning](<https://devfeed.tech/tags/learning.md>), [model](<https://devfeed.tech/tags/model.md>), [post-training](<https://devfeed.tech/tags/post-training.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [video](<https://devfeed.tech/tags/video.md>)

## AI overview

VIOS targets video-reasoning tokens where a model underuses visual evidence. It compares next-token distributions with and without video for a student and an answer-privileged teacher; their per-token reliance gap identifies positions for focused supervision. An EMA teacher, a contrastive objective, and a reliance-gap gate provide dense training signals. The method uses one on-policy rollout from one model, with no external rewards, judges, or larger teachers. The abstract reports improvements across 12 benchmarks and three model backbones, with over 9x fewer training steps, 60x fewer FLOPs, and 111x fewer generated tokens than contemporary video-reasoning systems.

## Source excerpt

arXiv:2611.11949v1 Announce Type: cross Abstract: Video reasoning hinges on a small number of visually decisive tokens embedded in otherwise largely language-driven chains of thought. Yet prevailing post-training methods such as GRPO overlook this structure, assigning coarse trajectory-level rewards while requiring multiple rollouts and often costly external judges. We introduce Visually-Aware On-Policy Self-Distillation (VIOS), a post-training framework that identifies where a model under-utilizes visual evidence and concentrates learning at those positions. Our key insight is that a model can audit its own visual reliance: the divergence between its next-token distributions with and without the video reveals how strongly each prediction depends on the visual input. Measuring this reliance for both a student and an answer-privileged teacher exposes a per-token reliance gap, identifying positions where the teacher exploits visual evidence that the student overlooks. VIOS translates this signal into dense, targeted supervision through three complementary components: an exponential moving average (EMA) teacher that co-evolves with the student, a contrastive objective that encourages predictions to depend on the video, and a reliance-gap gate that focuses learning on visually under-utilized tokens. Crucially, VIOS requires only a single on-policy rollout from a single model, without external rewards, judges, or larger teachers. Across 12 benchmarks spanning general, temporal, spatial, knowledge-intensive, and long-video understanding, VIOS consistently improves three model backbones and outperforms contemporary video-reasoning systems, while training in over 9x fewer steps, 60x fewer FLOPs, and 111x fewer generated tokens. Project page: https://vision.cs.utexas.edu/projects/VIOS/