Selm: Self-Evolving Library Management for Multi-Skill Agent Optimization

Key points

• Large language model (LLM) agents increasingly use skills as a lightweight way to acquire task-specific knowledge.1
• SELM jointly manages skill retrieval, updating, and lifecycle by maintaining skill confidence and utility signals.1
• It introduces utility-aware retrieval to select useful skills, a Map-Reduce framework to generate targeted updates from task-specific experience, and lifecycle controls to retain or remove skills based on their effectiveness.1
• We evaluate SELM on four agent benchmarks covering diverse tasks and execution environments, and SELM consistently outperforms in most settings.1

Sources (1)

• [1]Selm: Self-Evolving Library Management for Multi-Skill Agent Optimization
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 11:07 AM

  However, these methods often lack systematic management of evolved skills, making it difficult for agents to reuse and accumulate skills across domains, thus limiting performance on subsequent tasks.To reliably accumulate, reuse, and maintain useful skills across tasks, we present SELM (Self-Evolving Library Management), a framework that formulates skill evolution as library-level management.

  Large language model (LLM) agents increasingly use skills as a lightweight way to acquire task-specific knowledge.

Extractive summary: sentences quoted from the sources.