deepmind-research/functional_regularisation_for_continual_learning

Functional Regularisation for Continual Learning (FRCL) with Gaussian Processes

This notebook contains code to accompany the ICLR 2020 paper on Functional Regularisation for Continual Learning (FRCL).

The experiments in the paper can be reproduced by choosing the respective experimental protocol detailed in the publication. Default values are set to replicate results on Permuted MNIST.

If you find this code useful, please consider citing this paper:

@article{titsias2019functional,
  title={Functional regularisation for continual learning using gaussian processes},
  author={Titsias, Michalis K and Schwarz, Jonathan and Matthews, AG de G and Pascanu, Razvan and Teh, Yee Whye},
  journal={ICLR 2020},
  year={2020}
}