Reproduce “From Relevant Memories to Sufficient Evidence: Rethinking Memory Utilization in LLM Agents” 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
Shixuan Li, Wei Yang, Peiyu Zhang, Anzhe Cheng, Heng Ping, Xiaole Zhang, Jesse Thomason, Xuezhe Ma and 2 more
QEMem addresses failures in long-term conversational memory by assembling query-specific evidence rather than selecting relevant records alone. It searches multiple memory granularities, organizes findings into source-linked ViewCards, and uses estimated gaps to request more evidence. On LoCoMo, Claude Opus 4.5 reaches 91.43% LLM-judge accuracy, versus 83.83% for MemR³, across 1,540 questions; evaluation covers one benchmark and two Claude backbones.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.