Learning from Revision Consequences: Hindsight Meta-Experience Distillation for Self-Improving Agents

Key points

• As agents continuously improve by generating and revising Skills, the process that discovers and refines those Skills becomes a learnable object in its own right.1
• Task-Skills directly act on task execution, whereas Meta-Skills govern how agents discover and improve future Skills; their value therefore emerges through the subsequent search processes they induce.1
• Existing approaches improve Meta-Skills from observed raw Skill-search trajectories and branch outcomes.1
• Across three interactive agent benchmarks and both open-source and closed-source models, HMED consistently improves Skill discovery performance over strong baselines, shifting Meta-Skill learning beyond branch outcomes toward the consequences of changing the improvement process.1

Sources (1)

• [1]Learning from Revision Consequences: Hindsight Meta-Experience Distillation for Self-Improving Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 08:43 AM

  We introduce HMED (Hindsight Meta-Experience Distillation), a mechanism for constructing Meta-Experience for self-improving agents.

  As agents continuously improve by generating and revising Skills, the process that discovers and refines those Skills becomes a learnable object in its own right.

Extractive summary: sentences quoted from the sources.