Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

"My agent understands me better": Integrating Dynamic Human-like Memory Recall and Consolidation in LLM-Based Agents

Yuki Hou, Haruki Tamoto, Homei Miyashita

TL;DR

The paper presents a human-like memory architecture for LLM-based dialogue agents that enables autonomous, cue-triggered recall and dynamic memory consolidation to address temporal cognition limitations in transformers. It formalizes memory recall with an exponential-decay framework, where recall probability $p(t)$ is driven by memory relevance $r$ and a time-dependent decay $a$, adjusted over repeated recalls via $g_n$ and $S(t)$, and uses a recall threshold $k$ (≈0.86) to trigger retrieval. Memories are stored in a vector-database with content and temporal context, and are retrieved to influence prompts, enabling context-aware and personalized responses while keeping prompt length compact. Experimental results show statistically significant improvements in recall accuracy over Generative Agents across 10 tasks, supported by both quantitative loss analyses and qualitative dialogue examples, demonstrating the potential of human-like temporal cognition in AI agents and guiding future work on efficiency, behavior-shift detection, and domain generalization.

Abstract

In this study, we propose a novel human-like memory architecture designed for enhancing the cognitive abilities of large language model based dialogue agents. Our proposed architecture enables agents to autonomously recall memories necessary for response generation, effectively addressing a limitation in the temporal cognition of LLMs. We adopt the human memory cue recall as a trigger for accurate and efficient memory recall. Moreover, we developed a mathematical model that dynamically quantifies memory consolidation, considering factors such as contextual relevance, elapsed time, and recall frequency. The agent stores memories retrieved from the user's interaction history in a database that encapsulates each memory's content and temporal context. Thus, this strategic storage allows agents to recall specific memories and understand their significance to the user in a temporal context, similar to how humans recognize and recall past experiences.

"My agent understands me better": Integrating Dynamic Human-like Memory Recall and Consolidation in LLM-Based Agents

TL;DR

The paper presents a human-like memory architecture for LLM-based dialogue agents that enables autonomous, cue-triggered recall and dynamic memory consolidation to address temporal cognition limitations in transformers. It formalizes memory recall with an exponential-decay framework, where recall probability is driven by memory relevance and a time-dependent decay , adjusted over repeated recalls via and , and uses a recall threshold (≈0.86) to trigger retrieval. Memories are stored in a vector-database with content and temporal context, and are retrieved to influence prompts, enabling context-aware and personalized responses while keeping prompt length compact. Experimental results show statistically significant improvements in recall accuracy over Generative Agents across 10 tasks, supported by both quantitative loss analyses and qualitative dialogue examples, demonstrating the potential of human-like temporal cognition in AI agents and guiding future work on efficiency, behavior-shift detection, and domain generalization.

Abstract

In this study, we propose a novel human-like memory architecture designed for enhancing the cognitive abilities of large language model based dialogue agents. Our proposed architecture enables agents to autonomously recall memories necessary for response generation, effectively addressing a limitation in the temporal cognition of LLMs. We adopt the human memory cue recall as a trigger for accurate and efficient memory recall. Moreover, we developed a mathematical model that dynamically quantifies memory consolidation, considering factors such as contextual relevance, elapsed time, and recall frequency. The agent stores memories retrieved from the user's interaction history in a database that encapsulates each memory's content and temporal context. Thus, this strategic storage allows agents to recall specific memories and understand their significance to the user in a temporal context, similar to how humans recognize and recall past experiences.
Paper Structure (18 sections, 14 equations, 2 figures, 4 tables)

This paper contains 18 sections, 14 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Similarities Between LLMs and Human Memory
  4. Human-like Memory Processes in AI Agent
  5. Mathematical Models of Human Memory Processes
  6. LLM-based Autonomous Agents
  7. Architecture
  8. Model
  9. Memory Recall and Consolidation in Database Architecture
  10. Experiment
  11. Setup
  12. Analysis
  13. Memory Recall Accuracy
  14. Calculation of the Loss Function
  15. Response Generation Performance
  16. ...and 3 more sections

Figures (2)

  • Figure 1: (A) Decline in recall probability over time. The black curve ($r$=1, $g$=1) shows a rapid loss of recall, while the red curve ($r$=0.6, $g$=2) represents a slower forgetting rate. This difference indicates the challenge in designing dialogue agents that must distinguish between recent and distant events. (B) At time $t_0$, Event D is recalled by the user, and the model updates its temporal significance. This exemplifies how memory is reinforced through repetition, becoming less susceptible to forgetting at $t_0$+$t$.
  • Figure 2: (A) Comparison of Loss Values Between two Models Across Different Tasks (B) Verification of Significance of Results