/ROGO

This repo contains the codes for our work “Restricted orthogonal gradient projection for continual learning”.

Primary LanguagePythonMIT LicenseMIT

ROGO

This repo contains the codes for our work “Restricted orthogonal gradient projection for continual learning”.

Data Preparation

The dataset for PMNIST, CIFAR-100, and Mixture will be automatically downloaded. For the experiments on MiniImageNet, please download the train_data and test_data, and place them under the data folder.

Experiments

This repository currently contains experiments reported in the paper for Permuted MNIST, 10-split CIFAR-100, 20-split MiniImageNet and the Mixture datasets. All these experiments can be run using the following command:

bash run.sh

Citation

@article{YANG202398,
title = {Restricted orthogonal gradient projection for continual learning},
journal = {AI Open},
volume = {4},
pages = {98-110},
year = {2023},
issn = {2666-6510},
doi = {https://doi.org/10.1016/j.aiopen.2023.08.010},
url = {https://www.sciencedirect.com/science/article/pii/S2666651023000128},
author = {Zeyuan Yang and Zonghan Yang and Yichen Liu and Peng Li and Yang Liu},
}