Modular meta-learning
This addresses the need for more efficient learning in robotics and similar domains by enabling combinatorial generalization, though it appears incremental as it builds on existing meta-learning and modular approaches.
The paper tackles the problem of accelerating learning in prediction tasks with underlying structure, such as robotics, by proposing a strategy for learning neural network modules that can be combined in new ways to generalize to unseen tasks, achieving improved performance in two robotics-related problems.
Many prediction problems, such as those that arise in the context of robotics, have a simplifying underlying structure that, if known, could accelerate learning. In this paper, we present a strategy for learning a set of neural network modules that can be combined in different ways. We train different modular structures on a set of related tasks and generalize to new tasks by composing the learned modules in new ways. By reusing modules to generalize we achieve combinatorial generalization, akin to the "infinite use of finite means" displayed in language. Finally, we show this improves performance in two robotics-related problems.