Abstract: Highlights•Aiming to make classification models more stable.•Making a systematical empirical study on tangent loss.•Comparing with cross-entropy loss on various classification tasks.•Achieving better or competitive performance than cross-entropy loss.
Loading