We provide MATLAB codes to implement our method and other benchmarks used in our paper (https://arxiv.org/abs/1907.10595):
- "functions": This folder contains the necessary functions to run each metho, e.g. cost functions and their gradient for one and four layer network, quantizer, weghit initilization, etc.
- After loading with propoer dataset (MNIST or CIFAR-10), each method returns the corresponding training loss vs. training time.