Reproduce “Memory as Belief: Bayesian Evolution of Memory for Self-Evolving 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
Yiming Xu, Qian Yang, Zihan Chen, Yuefei Wu, Wenhan Zeng, Bin Shi, Bo Dong
BEAM models each memory’s usefulness as an uncertain belief conditioned on the current task, then updates it using execution feedback. It selects memories, synthesizes guidance, and uses leave-one-out counterfactuals to estimate contributions within combinations. Across five benchmark columns, its mean is 53.44 versus A-Mem’s 49.51 under DeepSeek-V4-Flash; it does not lead on every task, and evidence is limited to tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.