
PROCESS
Providing computing solutions for exascale challenges
This repository contains the development of the interpretability and perturbation robustness for the analysis of histopathology images by convolutional neural networks.
Recent work on Testing with Concept Activation Vectors (TCAV) proposed directional derivatives to quantify the in uence of user-defined concepts on the network output [1]. However, diagnostic concepts are often continuous measures that might be counter intuitive to describe by their presence or absence. In this paper, we extend TCAV from a classification problem to a regression problem by computing Regression Concept Vectors (RCVs). Instead of seeking a discriminator between two concepts, we seek the direction of greatest increase of the measures for a single continuous concept. We measure the relevance of a concept with bidirectional relevance scores, Br. The Br scores assume positive values when increasing values of the concept measures positively affect classification and negative in the opposite case.