Memory as Belief: Bayesian Evolution of Memory for Self-Evolving LLM Agents

Key points

• Large language model (LLM) agents increasingly rely on external memory to accumulate past experiences for continual adaptation.1
• Despite its importance, memory valuation remains largely underexplored, with existing methods relying on historical retrieval success rates that overlook the context dependency of memory usefulness and the varying reliability of estimates under different amounts of evidence.1
• Specifically, BEAM infers memory value beliefs from current task contexts and continuously calibrates them using execution feedback as empirical evidence, transforming memory from passive storage into an adaptive cognitive mechanism for continual self-evolution.1
• Extensive experiments across reasoning and code generation benchmarks demonstrate that BEAM consistently improves memory utilization and enables continual self-evolution of LLM agents.1

Sources (1)

• [1]Memory as Belief: Bayesian Evolution of Memory for Self-Evolving LLM Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 08:49 AM

  Inspired by the Bayesian brain hypothesis, we propose BEAM (Bayesian-style Evolutionary Agentic Memory), a framework that reformulates memory value evaluation as a parameterized cognitive inference process, where memory value is represented as a context-conditioned dynamic belief rather than a passive historical statistic.

  Large language model (LLM) agents increasingly rely on external memory to accumulate past experiences for continual adaptation.

Extractive summary: sentences quoted from the sources.