Believe Your Model: Distribution-Guided Confidence Calibration
This addresses the problem of improving answer selection reliability in reasoning models for AI applications, representing an incremental improvement over existing confidence-based methods.
The paper tackles the problem of unreliable confidence scores in Large Reasoning Models by proposing DistriVoting, which incorporates distributional priors alongside confidence during answer selection, and SelfStepConf to dynamically adjust inference. Experiments across 16 models and 5 benchmarks show it significantly outperforms state-of-the-art approaches.
Large Reasoning Models have demonstrated remarkable performance with the advancement of test-time scaling techniques, which enhances prediction accuracy by generating multiple candidate responses and selecting the most reliable answer. While prior work has analyzed that internal model signals like confidence scores can partly indicate response correctness and exhibit a distributional correlation with accuracy, such distributional information has not been fully utilized to guide answer selection. Motivated by this, we propose DistriVoting, which incorporates distributional priors as another signal alongside confidence during voting. Specifically, our method (1) first decomposes the mixed confidence distribution into positive and negative components using Gaussian Mixture Models, (2) then applies a reject filter based on positive/negative samples from them to mitigate overlap between the two distributions. Besides, to further alleviate the overlap from the perspective of distribution itself, we propose SelfStepConf, which uses step-level confidence to dynamically adjust inference process, increasing the separation between the two distributions to improve the reliability of confidences in voting. Experiments across 16 models and 5 benchmarks demonstrate that our method significantly outperforms state-of-the-art approaches.