Anomalous behaviour in loss-gradient based interpretability methods
This work highlights potential flaws in interpretability methods, which could mislead researchers and practitioners relying on these tools for model understanding.
The paper investigates loss-gradient attribution methods for interpreting deep learning models, finding that occluding parts of the input can sometimes improve performance on test datasets in sound and image recognition tasks.
Loss-gradients are used to interpret the decision making process of deep learning models. In this work, we evaluate loss-gradient based attribution methods by occluding parts of the input and comparing the performance of the occluded input to the original input. We observe that the occluded input has better performance than the original across the test dataset under certain conditions. Similar behaviour is observed in sound and image recognition tasks. We explore different loss-gradient attribution methods, occlusion levels and replacement values to explain the phenomenon of performance improvement under occlusion.