Layer-wise Relevance Propagation for Convolutional Neural Networks

Implementation of layer-wise relevance propagation on a convolutional neural network trained on the MNIST handwritten digit dataset. Example of heat maps for select layers for a sample input image of 7:

The corresponing relevance scores for every output class for the smaple input image of 7:

Example of heat maps for select layers for a sample input image of 8:

The corresponing relevance scores for every output class for the smaple input image of 8:
\