Underwhelming Generalization Improvements From Controlling Feature Attribution

Joseph D Viviano, Becks Simpson, Francis Dutil, Yoshua Bengio, Joseph Paul Cohen

Keywords: interpretability, medical, generalization, saliency

TL;DR: There is hope that one can diagnose and fix overfitting in classifiers by studying and guiding their saliency maps, but we developed multiple methods to do this well and only see a minor positive effect on generalization.

Abstract: Overfitting is a common issue in machine learning, which can arise when the model learns to predict class membership using convenient but spuriously-correlated image features instead of the true image features that denote a class. These are typically visualized using saliency maps. In some object classification tasks such as for medical images, one may have some images with masks, indicating a region of interest, i.e., which part of the image contains the most relevant information for the classification. We describe a simple method for taking advantage of such auxiliary labels, by training networks to ignore the distracting features which may be extracted outside of the region of interest, on the training images for which such masks are available. This mask information is only used during training and has an impact on generalization accuracy in a dataset-dependent way. We observe an underwhelming relationship between controlling saliency maps and improving generalization performance.

Code: https://github.com/bigtrellis2222/activmask

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:1910.00199/code)