Reproduce “Hierarchical Memory Orchestration for Personalized Persistent 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
Junming Liu, Yifei Sun, Weihua Cheng, Haodong Lei, Yuqi Li, Yirong Chen, Ding Wang
Hierarchical Memory Orchestration (HMO) is introduced to provide personalized and persistent long-term memory for intelligent agents, structuring interaction history into a three-tiered hierarchy for efficient recall. The framework achieved an 81.1% Recall@5 and 85.6% NDCG@5 on LongMemEval-S, demonstrating a 77.0% speedup in inference latency compared to exhaustive search with a marginal accuracy drop.
Key insights
- A multi-tiered hierarchical memory structure significantly prunes the search space, enabling faster and more precise retrieval of historical context for agents.
- Integrating an evolving, user-centric persona into memory prioritization ensures that agents maintain a personalized understanding, enhancing interaction fluidity and coherence.
- A hybrid approach to memory scoring and update, combining MLLM assessment, dynamic similarity alignment, and lazy updates, optimizes for both retrieval accuracy and computational efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.