Machine Learning and Neuroscience: Continual, Meta, and Reinforcement Learning

My doctoral work is focused on using understanding and inspiration from neuroscience to improve understanding and methods in machine learning. Of particular interest are continual learning - effectively learning tasks in sequence via consolidation and transfer, meta-learning - learning task distributions and/or learning to learn, and reinforcement learning - an interactive paradigm of machine learning in which agents make decisions in an environment to maximise a notion of reward. In this report I present two specific sub-projects: the re- aims to use neuromodulation in the brain as an inspiration for better exploratory reinforcement learning, the second investigates the interplay between task similarity and various preventative measures for catastrophic forgetting in a student-teacher continual learning framework.