Piyush-555/VCL-in-PyTorch

Variational Continual Learning in PyTorch

Re-Implementation of Variational Continual Learning by Nguyen, Li, Bui, and Turner (ICLR 2018).

SplitMNIST: VCL v/s Baselines

PermutedMNIST: VCL v/s Baselines

Vary beta (Weight coefficient for KL term)

Varying beta while keeping memory fixed at 200 samples per task.

Vary Memory

Varying memory while keeping beta fixed at 1.

Vary Memory (Fixed prior)

Varying memory while keeping prior fixed to Gaussian distribution.