Reproduce “From Personal to Collective: On the Role of Local and Global Memory in LLM Personalization” 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

Zehong Wang, Junlin Wu, ZHaoxuan Tan, Bolian Li, Xianrui Zhong, Zheli Liu, Qingkai Zeng

A new framework, LoGo, integrates personalized local memory with evolving collective global memory to enhance Large Language Model (LLM) personalization. It consistently outperforms prior methods, improving performance by up to 18.05% for cold-start users and increasing output diversity by up to 4.70% for highly active users, while demonstrating robustness to temporal shifts.

Key insights

  • Integrating collective global knowledge significantly improves LLM personalization for cold-start users by providing generalizable context and mitigates bias for active users by offering population-level regularization.
  • Dynamically evolving global memory, updated over temporal segments, is crucial for maintaining relevance and robustness in personalized LLM systems, adapting to shifting user interests.
  • A mediator module effectively reconciles user-specific preferences with broader collective patterns, leading to superior performance and diversity compared to relying solely on one memory type.

Summary by alphaXiv.

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