Reproduce “RippleMem: From Isolated Retrieval to Associative Recollection for Long-Term Agent Memory” 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
Jingbo Ji, Lingyi Li, Xilong Cheng, Yuhao Zhou, Wenji Zhang, Yuting Tan, Yunxiao Qin
RippleMem introduces an event-centric long-term memory system for LLM agents, shifting from isolated retrieval to adaptive, associative recollection. It leverages cue-rich episodic memory units and a sparse event graph to recover comprehensive, distributed evidence, achieving an F1 score of 52.49% on LoCoMo and 86.60% accuracy on LongMemEval-S, outperforming existing baselines.
Key insights
- Effective long-term memory for LLM agents requires moving beyond isolated retrieval to an adaptive, associative recollection process where initially retrieved memories serve as cues for finding additional, specifically missing evidence.
- Constructing memory as 'cue-addressable event memories' with explicit structured cues (participants, locations, time) and organizing them into a sparse, typed event graph facilitates targeted and efficient associative expansion.
- A planned, iterative recollection strategy that dynamically selects 'anchors' from current evidence and defines 'missing-support targets' significantly improves the completeness of retrieved evidence for complex queries.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.