TCAV (Testing with Concept Activation Vectors) is a well-known tool for interpreting deep learning models first proposed by Kim et. al: https://arxiv.org/abs/1711.11279. This project implements a TCAV class in XOR_TCAV.ipynb and uses it to gain a greater understanding of the capabilities and limitations of TCAV as a model interpretation tool.
The graphs folder contains scripts that create the Graphviz visualizations found in XOR_TCAV.ipynb.