Pre or Post-Softmax Scores in Gradient-based Attribution Methods, What is Best?
arXiv:2306.13197 · doi:10.1109/ICPRS58416.2023.10179032
Abstract
Gradient based attribution methods for neural networks working as classifiers use gradients of network scores. Here we discuss the practical differences between using gradients of pre-softmax scores versus post-softmax scores, and their respective advantages and disadvantages.
8 pages, 2 figures, 2023 IEEE 13th International Conference on Pattern Recognition Systems (ICPRS)