Deep Online Learning via Meta-Learning: Continual Adaptation for Model-Based RL

TL;DR

This paper introduces MOLe, a meta-learning approach enabling deep neural networks to perform continual online adaptation in non-stationary environments, particularly improving model-based reinforcement learning in dynamic real-world scenarios.

Contribution

The paper develops a meta-learning framework for online deep learning that effectively adapts to changing tasks and distributions, enhancing model-based RL performance.

Findings

MOLe outperforms prior methods in non-stationary environments.

Meta-training improves online adaptation of deep models.

Effective continual learning demonstrated in diverse RL tasks.

Abstract

Humans and animals can learn complex predictive models that allow them to accurately and reliably reason about real-world phenomena, and they can adapt such models extremely quickly in the face of unexpected changes. Deep neural network models allow us to represent very complex functions, but lack this capacity for rapid online adaptation. The goal in this paper is to develop a method for continual online learning from an incoming stream of data, using deep neural network models. We formulate an online learning procedure that uses stochastic gradient descent to update model parameters, and an expectation maximization algorithm with a Chinese restaurant process prior to develop and maintain a mixture of models to handle non-stationary task distributions. This allows for all models to be adapted as necessary, with new models instantiated for task changes and old models recalled when previously seen tasks are encountered again. Furthermore, we observe that meta-learning can be used to meta-train a model such that this direct online adaptation with SGD is effective, which is otherwise not the case for large function approximators. In this work, we apply our meta-learning for online learning (MOLe) approach to model-based reinforcement learning, where adapting the predictive model is critical for control; we demonstrate that MOLe outperforms alternative prior methods, and enables effective continuous adaptation in non-stationary task distributions such as varying terrains, motor failures, and unexpected disturbances.