A Meta-MDP Approach to Exploration for Lifelong Reinforcement Learning

By Francisco M. Garcia, Philip S. Thomas. 2019

46 citations recorded by the source corpus.

In this paper we consider the problem of how a reinforcement learning agent that is tasked with solving a sequence of reinforcement learning problems (Markov decision processes) can use knowledge acquired early in its lifetime to improve its ability to solve new problems. Specifically, we focus on the question of how the agent should explore when faced with a new environment. We show that the search for an optimal exploration strategy can be formulated as a reinforcement learning problem itself, albeit with a different timescale. We conclude with experiments that show the benefits of optimizing an exploration strategy using our proposed approach.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Lifelong Policy Gradient Learning of Factored Policies for Faster Training Without Forgetting (2020, background): “Other approaches store experiences in memory for future replay [18, 30] or use a separate model for each task [31, 14].”
Routing Networks and the Challenges of Modular and Compositional Computation (2019, methodology): “For example, it may be useful to consider algorithms that have separate exploitation and exploration policies, along the lines of Garcia and Thomas [2019], or algorithms that can incorporate more complex exploration mechanics into reparameterization techniques.”
Deep reinforcement learning: a survey (2020, background): “The meta-MDP approach to exploration (Garcia and Thomas, 2019) models the problem of searching for an optimal exploration policy as an MDP, and separates the policies into an exploration policy and an exploitation policy to achieve lifelong learning.”
How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Composition (2022, methodology): “One such method leverages shared knowledge in the form of a metamodel that informs exploration strategies to task-specific models, resulting in linear growth of the model parameters (Garcia and Thomas, 2019).”
Human-Timescale Adaptation in an Open-Ended Task Space (2023, methodology): “either rely on high-variance Monte Carlo returns (Garcia and Thomas, 2019; Stadie et al., 2018; Vuorio et al., 2021) or history-dependent estimators (Zheng et al.”
Reactive Exploration to Cope with Non-Stationarity in Lifelong Reinforcement Learning (2022, background): “Garcia & Thomas (2019) address the problem of optimizing the exploration strategy in a meta-MDP.”