AI RESEARCH

Does Visual Token Pruning Improve Calibration? An Empirical Study on Confidence in MLLMs

arXiv CS.CV

ArXi:2604.12035v1 Announce Type: new Visual token pruning is a widely used strategy for efficient inference in multimodal large language models (MLLMs), but existing work mainly evaluates it with task accuracy. In this paper, we study how visual token pruning affects model calibration, that is, whether predicted confidence matches actual correctness.