Bayesian continual learning

Work in progress.

This project explores the possibility of using a hypernetwork controller, trained through reinforcement learning, to infer the correct downstream network to use for the purposes of continual learning.