Revisiting Self-Consistency from Dynamic Distributional Alignment Perspective on Answer Aggregation

TL;DR

This paper reinterprets self-consistency in answer aggregation as a dynamic distributional alignment problem, proposing a confidence-driven temperature calibration to improve reasoning performance with limited samples.

Contribution

It introduces a novel perspective on self-consistency as a dynamic alignment problem and proposes a temperature calibration mechanism to enhance answer aggregation.

Findings

Outperforms fixed-diversity baselines in mathematical reasoning tasks.

Improves average and best-case performance across different initial temperatures.

Does not require additional data or modules.

Abstract

Self-consistency improves reasoning by aggregating diverse stochastic samples, yet the dynamics behind its efficacy remain underexplored. We reframe self-consistency as a dynamic distributional alignment problem, revealing that decoding temperature not only governs sampling randomness but also actively shapes the latent answer distribution. Given that high temperatures require prohibitively large sample sizes to stabilize, while low temperatures risk amplifying biases, we propose a confidence-driven mechanism that dynamically calibrates temperature: sharpening the sampling distribution under uncertainty to align with high-probability modes, and promoting exploration when confidence is high. Experiments on mathematical reasoning tasks show this approach outperforms fixed-diversity baselines under limited samples, improving both average and best-case performance across varying initial temperatures without additional data or modules. This establishes self-consistency as a synchronization challenge between sampling dynamics and evolving answer distributions.