Linking Perception, Confidence and Accuracy in MLLMs

TL;DR

This paper identifies confidence miscalibration in Multi-modal Large Language Models and introduces a novel framework with confidence-based training and test-time scaling to improve perceptual sensitivity and overall performance.

Contribution

It proposes Confidence-Driven Reinforcement Learning and Confidence-Aware Test-Time Scaling to calibrate confidence and enhance MLLMs, achieving state-of-the-art results across multiple benchmarks.

Findings

Severe confidence miscalibration in MLLMs uncovered.

Proposed methods improve calibration and accuracy by 8.8%.

Framework achieves consistent gains across four benchmarks.

Abstract

Recent advances in Multi-modal Large Language Models (MLLMs) have predominantly focused on enhancing visual perception to improve accuracy. However, a critical question remains unexplored: Do models know when they do not know? Through a probing experiment, we reveal a severe confidence miscalibration problem in MLLMs. To address this, we propose Confidence-Driven Reinforcement Learning (CDRL), which uses original-noise image pairs and a novel confidence-based reward to enhance perceptual sensitivity and robustly calibrate the model's confidence. Beyond training benefits, calibrated confidence enables more effective test-time scaling as a free lunch. We further propose Confidence-Aware Test-Time Scaling (CA-TTS), which dynamically coordinates Self-Consistency, Self-Reflection, and Visual Self-Check modules guided by confidence signals. An Expert Model acts in multiple roles (e.g., Planner, Critic, Voter) to schedule these modules and provide external verification. Our integrated framework establishes new state-of-the-art results with consistent 8.8% gains across four benchmarks. More ablation studies demonstrate the effectiveness of each module and scaling superiority.