HiMeS: Hippocampus-inspired Memory System for Personalized AI Assistants

TL;DR

HiMeS introduces a hippocampus-inspired memory system for personalized AI assistants, combining short-term and long-term memory modules to improve user-specific interactions and question-answering performance.

Contribution

The paper presents a novel memory architecture inspired by hippocampus-neocortex mechanisms, integrating reinforcement learning-based short-term memory and partitioned long-term memory for personalized AI.

Findings

Outperforms baseline in question-answering quality on real-world data

Both memory modules are essential for optimal performance

Ablation studies validate the system's components and design choices

Abstract

Large language models (LLMs) power many interactive systems such as chatbots, customer-service agents, and personal assistants. In knowledge-intensive scenarios requiring user-specific personalization, conventional retrieval-augmented generation (RAG) pipelines exhibit limited memory capacity and insufficient coordination between retrieval mechanisms and user-specific conversational history, leading to redundant clarification, irrelevant documents, and degraded user experience. Inspired by the hippocampus-neocortex memory mechanism, we propose HiMeS, an AI-assistant architecture that fuses short-term and long-term memory. Our contributions are fourfold: (1) A short-term memory extractor is trained end-to-end with reinforcement learning to compress recent dialogue and proactively pre-retrieve documents from the knowledge base, emulating the cooperative interaction between the hippocampus and prefrontal cortex. (2) A partitioned long-term memory network stores user-specific information and re-ranks retrieved documents, simulating distributed cortical storage and memory reactivation. (3) On a real-world industrial dataset, HiMeS significantly outperforms a cascaded RAG baseline on question-answering quality. (4) Ablation studies confirm the necessity of both memory modules and suggest a practical path toward more reliable, context-aware, user-customized LLM-based assistants.