Unhackable Temporal Rewarding for Scalable Video MLLMs

TL;DR

This paper identifies the problem of temporal hacking in video MLLMs, introduces a theoretical framework and a new reward method to improve temporal understanding, and demonstrates significant performance gains.

Contribution

It establishes a theory of temporal hacking, proposes the UTR framework to prevent it, and introduces TPL as a metric for temporal modeling quality.

Findings

TPL correlates with frame activation patterns

UTR effectively counters temporal hacking

UTR improves video comprehension performance

Abstract

In the pursuit of superior video-processing MLLMs, we have encountered a perplexing paradox: the "anti-scaling law", where more data and larger models lead to worse performance. This study unmasks the culprit: "temporal hacking", a phenomenon where models shortcut by fixating on select frames, missing the full video narrative. In this work, we systematically establish a comprehensive theory of temporal hacking, defining it from a reinforcement learning perspective, introducing the Temporal Perplexity (TPL) score to assess this misalignment, and proposing the Unhackable Temporal Rewarding (UTR) framework to mitigate the temporal hacking. Both theoretically and empirically, TPL proves to be a reliable indicator of temporal modeling quality, correlating strongly with frame activation patterns. Extensive experiments reveal that UTR not only counters temporal hacking but significantly elevates video comprehension capabilities. This work not only advances video-AI systems but also illuminates the critical importance of aligning proxy rewards with true objectives in MLLM development.