Introducing Symmetries to Black Box Meta Reinforcement Learning. (arXiv:2109.10781v2 [cs.LG] UPDATED)

Meta reinforcement learning (RL) attempts to discover new RL algorithms
automatically from environment interaction. In so-called black-box approaches,
the policy and the learning algorithm are jointly represented by a single
neural network. These methods are very flexible, but they tend to underperform
in terms of generalisation to new, unseen environments. In this paper, we
explore the role of symmetries in meta-generalisation. We show that a recent
successful meta RL approach that meta-learns an objective for
backpropagation-based learning exhibits certain symmetries …

arxiv black box learning meta reinforcement reinforcement learning