How Much Evidence Should a Coding Agent's Self-Correction Carry? Adaptive Dirichlet Evidence for Self-Distillation

Key points

• A correction's learning weight should reflect both the transitions supported by its executions and the amount of evidence behind that support.1
• We introduce Effective-Evidence Self-Distillation (EESD), which represents these quantities separately.1
• Normalized execution relevance determines relative transition support and an effective pseudo-count mass; a Dirichlet posterior then produces an uncertainty-penalized weight for KL-anchored correction learning.1
• These results show how separating evidence support from evidence mass changes probability estimation and correction learning in coding agents.1

Sources (1)

• [1]How Much Evidence Should a Coding Agent's Self-Correction Carry? Adaptive Dirichlet Evidence for Self-Distillation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 03:17 PM

  Execution feedback lets coding agents revise programs and learn from their own corrections.

  A correction's learning weight should reflect both the transitions supported by its executions and the amount of evidence behind that support.

Extractive summary: sentences quoted from the sources.