xiaogaogaoxiao/BayesBiNN

Code for the paper "Training Binary Neural Networks with Bayesian Learning Rule

Code for paper "Training Binary Neural Networks using the Bayesian Learning Rule". https://arxiv.org/abs/2002.10778

1. Synthetic data

synthetic_data.ipynb

2. Image classification

BayesBiNN method:

MNIST: python main_mnist.py --model MLPBinaryConnect --optim BayesBiNN

Cifar10: python main_cifar10.py --model MLPBinaryConnect --optim BayesBiNN

Cifar100: python main_cifar100.py --model MLPBinaryConnect --optim BayesBiNN

STE-Adam method:

MNIST: python main_mnist.py --model MLPBinaryConnect_STE --optim STE

Cifar10: python main_cifar10.py --model MLPBinaryConnect_STE --optim STE

Cifar100: python main_cifar100.py --model MLPBinaryConnect_STE --optim STE

Full-Precision Adam:

MNIST: python main_mnist.py --model MLPBinaryConnect --optim Adam

Cifar10: python main_cifar10.py --model MLPBinaryConnect --optim Adam

Cifar100: python main_cifar100.py --model MLPBinaryConnect --optim Adam

3. Continual Learning

python main_permute_mnist_CL.py