Besides Brain-Inspired Replay (BI-R), we have currently implemented simple fine-tuning (None), Synaptic Intelligence (SI) and Learning without Forgetting (LwF) as baseline methods.
To run each of these methods, use the following:
- None:
python train_cifar100_incremental - LwF:
python train_cifar100_incremental --lwf - SI:
python train_cifar100_incremental --si - BI-R:
python train_cifar100_incremental --bir - BI-R + SI:
python train_cifar100_incremental --bir --si
Log-files for evaluating performance should be produced by running any of the above commands, but we are still working on evaluating these log-files using the l2metrics-environment.