Visualizing and Generalizing Integrated Attributions

Ethan Payne, David Patrick, Amanda S. Fernandez

Abstract: Explainability and attribution for deep neural networks remains an open area of study due to the importance of adequately interpreting the behavior of such ubiquitous learning models. The method of expected gradients [10] reduced the baseline dependence of integrated gradients [27] and allowed for improved interpretability of attributions as representative of the broader gradient landscape, however both methods are visualized using an ambiguous transformation which obscures attribution information and neglects to distinguish between color channels. While expected gradients takes an expectation over the entire dataset, this is only one possible domain in which an explanation can be contextualized. In order to generalize the larger family of attribution methods containing integrated gradients and expected gradients, we instead frame each attribution as a volume integral over a set of interest within the input space, allowing for new levels of specificity and revealing novel sources of attribution information. Additionally, we demonstrate these new unique sources of feature attribution information using a refined visualization method which allows for both signed and unsigned attributions to be visually salient for each color channel. This new formulation provides a framework for developing and explaining a much broader family of attribution measures, and for computing attributions relevant to diverse contexts such as local and non-local neighborhoods. We evaluate our novel family of attribution measures and our improved visualization method using qualitative and quantitative approaches with the CIFAR10 and ImageNet datasets and the Quantus XAI library.