Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training

Key points

• Large language models have achieved remarkable advances in complex reasoning.1
• However, scaling reasoning post-training with ever-larger corpora can substantially increase training cost without proportional performance gains.1
• In this study, we therefore ask: how far can reasoning supervision be compressed without sacrificing capability?1
• Specifically, rather than treating a teacher-generated chain-of-thought (CoT) as a single realized path, we construct uncertainty-profiled reasoning supervision by building Monte Carlo reasoning trees around intermediate reasoning states to expose process-level uncertainty.1

Sources (2)

• [1]Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 03:26 AM

  To this end, we propose an uncertainty-guided progressive contraction framework that concentrates supervision and optimization on the reasoning trajectories most valuable for learning.

  Large language models have achieved remarkable advances in complex reasoning.

• [2]Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training
  Hugging Face Daily Papers · Oct 9, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.