VideoLoop: Looped Working Memory Against Semantic Thrashing in Long-Form Video Agents
VideoLoop rewrites bounded memory so long-video agents avoid semantic thrashing, gaining 4.2 points.
VideoLoop targets semantic thrashing in long-form video agents, where growing append-only memory causes attention to key evidence to collapse. An outer reasoning loop is paired with an inner loop that retrieves past artifacts and rewrites a bounded working memory. It improves four LVLM backbones by 4.2 points on average on VideoMME long; with Gemini 3.1 Pro it scores 88.3% on VideoMME long, 88.8% on VideoMMMU, and 80.9% on LongVideoBench long.
- Append-only memory cannot discard noise without a rewrite operator.
- An inner loop rewrites bounded memory from an unbounded artifact store.
- Average gain is 4.2 points on VideoMME long across four LVLM backbones.
- A context-only judge scores 81.1% versus 60.9% for append-only memory.
- With Gemini 3.1 Pro it reaches 88.3% on VideoMME long.
- AI models
- VideoLoopGemini 3.1 Pro
From the original article
1 min read at huggingface.co
Long-form video understanding requires multimodal agents to iteratively gather evidence over many reasoning steps. However, most existing agentic methods suffer from semantic thrashing: as append-only working memory grows, attention to key evidence collapses, and the agent loses access to what it has already found.
By Jinfa Huang, Jianming Xu, Jingyang Lin · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.