This project aims at classification continual learning problems using Energy-Based Models. Mainly based on our paper Energy-Based Models for Continual Learning.
-
[Code] This code is the basic version of our paper. We will release the final version soon.
The current version of the code has been tested with:
pytorch 1.4.0
torchvision 0.2.1
sh scripts/boundary_aware/train_ebm_splitmnist.sh
sh scripts/boundary_aware/train_sbc_splitmnist.sh
sh scripts/boundary_aware/train_ebm_permmnist.sh
sh scripts/boundary_aware/train_sbc_permmnist.sh
sh scripts/boundary_aware/train_ebm_cifar10.sh
sh scripts/boundary_aware/train_sbc_cifar10.sh
sh scripts/boundary_aware/train_ebm_cifar100.sh
sh scripts/boundary_aware/train_sbc_cifar100.sh
sh scripts/boundary_agnostic/train_ebm_splitmnist.sh
sh scripts/boundary_agnostic/train_sbc_splitmnist.sh
sh scripts/boundary_agnostic/train_ebm_permmnist.sh
sh scripts/boundary_agnostic/train_sbc_permmnist.sh
sh scripts/boundary_agnostic/train_ebm_cifar10.sh
sh scripts/boundary_agnostic/train_sbc_cifar10.sh
sh scripts/boundary_agnostic/train_ebm_cifar100.sh
sh scripts/boundary_agnostic/train_sbc_cifar100.sh
Parts of the code were based on the implementation of https://github.com/GMvandeVen/continual-learning.
Please consider citing our papers if you use this code in your research:
@article{li2020energy,
title={Energy-Based Models for Continual Learning},
author={Li, Shuang and Du, Yilun and van de Ven, Gido M and Torralba, Antonio and Mordatch, Igor},
journal={arXiv preprint arXiv:2011.12216},
year={2020}
}