The Lottery Ticket Hypothesis

This repository is an implementation of LT

In this repo, there are two settings

1. On MNIST, using simple MLP (mlp_mnist)
2. On CIFAR10, using CNN (cnn_cifar)

How to use?

Default settings are written in configure.py You can change learning rate, pruned_ratio (i.e., how many parameters are pruned in total), and iteration (gradually prune parameters).

Example

Please enter the following command in each file.

---

python3 train.py --epochs 10 --lr 1.2e-3 --iteration 10 --ratio 0.95

---

Results are automatically saved in csv format and png (for ploting).