Figure 10: We are optimizing VSML RNNs to implement neural forwardcomputation suchthat for different inputs and weights a tanh-activated multiplicative interaction is produced (left), with different lines for differentw. Next, we use a deep network and provide intermediate errors by a ground truth network. Finally, we remove intermediate errors and use the RNN's intermediate predictions that are now close to the ground truth. All 6meta test tasks are unseen. Thebottom plot shows the same dataset processed by SGD with Adam which learns significantly slower by followingthegradient. those enabled.

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 (specifically the reuse of the learning rule, and invariance to input and output permutations) that are not present in typical black-box meta RL systems. We hypothesise that these symmetries can play an important role in meta-generalisation. Building off recent work in black-box supervised meta learning, we develop a black-box meta RL system that exhibits these same symmetries. We show through careful experimentation that incorporating these symmetries can lead to algorithms with a greater ability to generalise to unseen action & observation spaces, tasks, and environments.