Принято на NeurIPS 2022
На основе https://github.com/aimagelab/mammoth.
@inproceedings{bonicelli2022effectiveness,
title={On the Effectiveness of Lipschitz-Driven Rehearsal in Continual Learning},
author={Bonicelli, Lorenzo and Boschini, Matteo and Porrello, Angelo and Spampinato, Concetto and Calderara, Simone},
booktitle = {Advances in Neural Information Processing Systems 35},
year={2022}
}
Команда:
python utils/main.py --dataset=seq-cifar100 --model=er_ace_lipschitz --n_epochs=50 --buffer_size=500 --lr=0.1 --load_cp=checkpoints/erace_pret_on_tinyr.pth --pre_epochs=200 --datasetS=tinyimgR --non_verbose