Linking Perception, Confidence and Accuracy in MLLMs
This addresses a critical reliability issue for users of MLLMs by improving model self-awareness and accuracy, though it is incremental as it builds on existing perception-focused advancements.
The paper tackles the problem of confidence miscalibration in Multi-modal Large Language Models (MLLMs), proposing a framework that integrates Confidence-Driven Reinforcement Learning and Confidence-Aware Test-Time Scaling to achieve consistent 8.8% accuracy gains across four benchmarks.
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.