Replicating MNIST digit classification acc results of Binarized Neural Networks: Training Neural Networks with Weights and Activations Constrained to+1or−1 in tensorflow
Usage: python3 main_mnist.py
Current Acc 98.86% at 1000 Epochs.
Training took ~4 Hours on an NVIDIA Tesla V100-SXM2 16GB GPU.
GPU utiliziation using nvidia-smi: