Abstract: We study model confidence calibration in class-incremental learning, where models learn from sequential tasks with different class sets. While existing works primarily focus on accuracy, ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback