Continual learning with a Bayesian CNN

Instructions

main_continual.py is the main file. Run it each time with the subsequent hyperparameters configurations:

1. pretrained = False will give you weights_1.pkl (everything else as it is by default)
2. pretrained = True, and task = 2 will give you weights_2.pkl to evaluate forward transfer
3. is_training = False, pretrained = True, task = 2 to evaluate backward transfer, i.e. ability to overcome catastrophic forgetting
4. pretrained = True, task = 3 will give you weights_3.pkl
5. is_training = False, pretrained = True, task = 3
6. (and so on for as many tasks as you want)

how we transfer parameters

Results

Results on permutations on MNIST

Results on similar CIFAR-100 classes ("leopards", "tigers", "lions")

Results on dissimilar CIFAR-100 classes ("leopards", "palms", "bicycles")

Bayesian CNN repository