Learning from Contrasts: Synthesizing Reasoning Paths from Diverse Search Trajectories

TL;DR

This paper introduces CRPS, a contrastive synthesis framework for reasoning paths that improves model performance and generalization by leveraging differences between search trajectories.

Contribution

CRPS transforms supervision from filtering to synthesis, extracting strategic insights from contrasting search paths to create more effective training data.

Findings

Models trained on 60K CRPS examples match/exceed 590K baseline performance.

CRPS enhances out-of-domain generalization.

Contrastive learning yields more transferable reasoning.

Abstract

Monte Carlo Tree Search (MCTS) has been widely used for automated reasoning data exploration, but current supervision extraction methods remain inefficient. Standard approaches retain only the single highest-reward trajectory, discarding the comparative signals present in the many explored paths. Here we introduce \textbf{Contrastive Reasoning Path Synthesis (CRPS)}, a framework that transforms supervision extraction from a filtering process into a synthesis procedure. CRPS uses a structured reflective process to analyze the differences between high- and low-quality search trajectories, extracting explicit information about strategic pivots and local failure modes. These insights guide the synthesis of reasoning chains that incorporate success patterns while avoiding identified pitfalls. We show empirically that models fine-tuned on just 60K CRPS-synthesized examples match or exceed the performance of baselines trained on 590K examples derived from standard rejection sampling, a 20$\times$ reduction in dataset size. Furthermore, CRPS improves generalization on out-of-domain benchmarks, demonstrating that learning from the contrast between success and failure produces more transferable reasoning capabilities than learning from success alone.