Although deep neural networks have made significant progress in confidence calibration, conventional methods like temperature scaling and histogram binning face notable limitations in dynamic ...