Safe Learning Under Irreversible Dynamics via Asking for Help

TL;DR

This paper introduces a safe learning algorithm that leverages mentor assistance and knowledge transfer to achieve low regret in irreversible Markov Decision Processes, enabling effective learning in high-stakes environments.

Contribution

It presents the first formal proof that an agent can learn effectively and safely in irreversible environments by asking for help and transferring knowledge, with sublinear regret and mentor queries.

Findings

Regret and mentor queries are both sublinear in the time horizon.

The approach applies to any Markov Decision Process, including irreversible ones.

Provides a sequence of reductions of independent interest.

Abstract

Most learning algorithms with formal regret guarantees essentially rely on trying all possible behaviors, which is problematic when some errors cannot be recovered from. Instead, we allow the learning agent to ask for help from a mentor and to transfer knowledge between similar states. We show that this combination enables the agent to learn both safely and effectively. Under standard online learning assumptions, we provide an algorithm whose regret and number of mentor queries are both sublinear in the time horizon for any Markov Decision Process (MDP), including MDPs with irreversible dynamics. Our proof involves a sequence of three reductions which may be of independent interest. Conceptually, our result may be the first formal proof that it is possible for an agent to obtain high reward while becoming self-sufficient in an unknown, unbounded, and high-stakes environment without resets.