Reproduce “Latent Multimodal Memory Reasoning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Cai Ke, Xinghao Chen, Yuanchen Bei, Yuni Lai, Keyu Chen, Ruizhi Qiao, Siyu An, Junnan Dong and 2 more

LMMR organizes text and visual evidence from dialogue turns into a query-conditioned memory before generating answers. It links shared concepts to temporal relations while retaining visual tokens for fine-grained detail. On Mem-Gallery, using Qwen2.5-VL-7B, it achieves overall token-level F1 of 79.10, versus 69.66 for MuRAG; this benchmark-wide result does not establish which component causes the difference.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.