Variational Continual Learning (VCL)

An implementation of the Variational Continual Learning (VCL) algorithms proposed by Nguyen, Li, Bui, and Turner (ICLR 2018).

@inproceedings{nguyen2018variational,
  title = {Variational Continual Learning},
  author = {Nguyen, Cuong V. and Li, Yingzhen and Bui, Thang D. and Turner, Richard E.},
  booktitle = {International Conference on Learning Representations},
  year = {2018}
}

To run the Permuted MNIST experiment:

python run_permuted.py

To run the Split MNIST experiment:

python run_split.py

To run the Generative MNIST experiment:

python sandbox/torch_generative.py

Requirements:

• Torch 1.0
• Python 3.6

Results

VCL in Deep discriminative models

Permuted MNIST

Split MNIST ... with Variational Generative Replay (VGR):

Results

VCL in Deep Generative models