This repository is an appendix to the master's thesis mentionned above. It contains the code used for the experimental analysis of the weight sharing approach to improve fault tolerance of memristor crossbar-based neural network.
Neural Network for MNIST.ipynb: It contains the code for a simple neural network consisting of two linear layers for digit recognition of MNIST dataset. Layers vs Single Layer.ipynb: This contains the code for implementating the weights having different fault percentages between 0 to 14% for Single layer memristor crossbar and triple layer memristor crossbar and stores the result in CSV format.