RESEARCH · RESEARCH · #1841
Uncertainty-Guided Progressive Contraction for efficient reasoning post-training (arXiv:2610.12933v1)
The paper proposes an uncertainty-guided progressive contraction framework that compresses reasoning supervision and focuses optimization on high-value reasoning trajectories. It builds Monte Carlo reasoning trees around intermediate CoT states to profile uncertainty, first contracts a teacher-informed training pool and then performs learner-aware online routing during SFT, claiming up to ~450x less training exposure (as little as 0.22% of full-data SFT) while maintaining or improving reasoning across benchmarks and diverse LLM families.
KEY POINTS
- The paper proposes an uncertainty-guided progressive contraction framework that compresses reasoning supervision and focuses optimization on high-value reasoning trajectories.
- It builds Monte Carlo reasoning trees around intermediate CoT states to profile uncertainty, first contracts a teacher-informed training pool and then performs learner-aware online routing during SFT, claiming up to ~450x less training exposure (as little as 0.22% of full-data SFT) while maintaining or improving reasoning across benchmarks and diverse LLM families.
- If reproducible, the method could sharply reduce SFT compute for reasoning tasks and enable faster or cheaper deployment of reasoning-capable models across heterogeneous student and teacher setups.
WHY IT MATTERS
If reproducible, the method could sharply reduce SFT compute for reasoning tasks and enable faster or cheaper deployment of reasoning-capable models across heterogeneous student and teacher setups.