This repository contains a Pytorch implementation of the paper Multi-prize Lottery Ticket Hypothesis.

pip3 install -r requirements.txt

python3 main.py

• --lr : Learning rate
  • Default : 1.2e-3
• --epochs : Number of cycle of pruning that should be done.
  • Default : 50
• --test_freq : Frequency for Validation
  • Default : 50
• --batch_size : Batch size
  • Default : 60
• --dataset : Choice of dataset
  • Options : mnist, cifar10
  • Default : mnist
• --arch_type : Type of architecture
  • Options : fc1 - Simple fully connected network, lenet5 - LeNet5, AlexNet - AlexNet, resnet18 - Resnet18, vgg16 - VGG16
  • Default : fc1
• --prune_percent : Percentage of weight to be pruned after each cycle.
  • Default : 5
• --mini_batch : Experiment on mini-batch
  • Default : False
• --score : Using score matrix to determine the pruning mask
  • Default : False
• --binarize : Model binarization
  • Default : False

The-Lottery-Ticket-Hypothesis-Binary-Neural-Networks-Pruning
├── models
│   ├── cifar10
│   │   ├── AlexNet.py
│   │   ├── fc1.py
│   │   ├── LeNet5.py
│   │   ├── resnet.py
│   │   ├── SmallVGG.py
│   │   └── vgg.py
│   └── mnist
│       ├── AlexNet.py
│       ├── fc1.py
│       ├── LeNet5.py
│       ├── resnet.py
│       ├── SmallVGG.py
│       └── vgg.py
├── dumps
├── main.py
├── plots
├── README.md
├── requirements.txt
├── saves
└── utils.py