Identifying Spurious Correlations and Correcting them with an Explanation-based Learning
This addresses the issue of model trustworthiness for AI practitioners by refining trained models, though it appears incremental as it builds on existing explanation-based methods.
The paper tackles the problem of identifying spurious correlations in trained image classification models by applying image-level perturbations and monitoring prediction certainty changes, demonstrating overdependence on spurious regions in a synthetic dataset and correcting it with an explanation-based learning approach.
Identifying spurious correlations learned by a trained model is at the core of refining a trained model and building a trustworthy model. We present a simple method to identify spurious correlations that have been learned by a model trained for image classification problems. We apply image-level perturbations and monitor changes in certainties of predictions made using the trained model. We demonstrate this approach using an image classification dataset that contains images with synthetically generated spurious regions and show that the trained model was overdependent on spurious regions. Moreover, we remove the learned spurious correlations with an explanation based learning approach.