Link to paper: Improving and Understanding Variational Continual Learning
Requirement: Tensorflow 1.12.0.
To run the Permuted MNIST experiment:
python run_permuted.py
To run the Split MNIST experiment:
python run_multihead_split.py
To plot weights from trained models on both experiments:
python plot_weights.py
The printed results are matrices where each row i contains the test set accuracies on all previously observed tasks after seeing task i. The average accuracy will also be plotted and saved to model_storage/. They should be similar to the figures in Appendix A in the paper.