Deep neural networks, including large language models, have achieved remarkable performance across various tasks. However, they are prone to overconfidence during training or fine-tuning. In this work, we observe a consistent phenomenon across different models that the early model is better calibrated, while later trai...
Yu-Tong Xie, Jia-Wei Tang, Zheng-Lin Hua et al.· 0 citations
Endo-C6, a compact corruption benchmark of six endoscopy-realistic perturbations evaluated at a fixed high severity, is introduced and RobustEndoCLIP, obtained by few-shot parameter-efficient tuning with VeRA, outperforming existing TVLM baselines are presented.
Darakshan Rashid, Raza Imam, Ufaq Khan et al.· 0 citations
The underlying causes of miscalibration in VLMs and Medical-VLMs are investigated and a simple train-time calibration method, Multi-Class Margin (MCM) regularization, is proposed, which achieves lowest ECE on 10 out of 12 settings in in-domain and remains competitive under domain shifts.
MoBE is proposed: a fully optimization-free framework that performs dynamic expert selection and adaptation at test time that combines entropy-guided dynamic routing in MoE settings with expert-wise Bayesian adaptation, enabling experts to update their confidence and adapt online without gradient updates.
