Work in progress.

This project explores the use of a hypernetwork controller, trained through reinforcement learning, to infer which downstream subnetwork to use for purposes of continual learning.