English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In online reinforcement learning (online RL), balancing exploration andexploitation is crucial for finding an optimal policy in a sample-efficientway. To achieve this, existing sample-efficient online RL algorithms typicallyconsist of three components: estimation, planning, and exploration. However, inorder to cope with general function approximators, most of them involveimpractical algorithmic components to incentivize exploration, such asoptimization within data-dependent level-sets or complicated samplingprocedures. To address this challenge, we propose an easy-to-implement RLframework called \textit{Maximize to Explore} (\texttt{MEX}), which only needsto optimize \emph{unconstrainedly} a single objective that integrates theestimation and planning components while balancing exploration and exploitationautomatically. Theoretically, we prove that \texttt{MEX} achieves a sublinearregret with general function approximations for Markov decision processes (MDP)and is further extendable to two-player zero-sum Markov games (MG). Meanwhile,we adapt deep RL baselines to design practical versions of \texttt{MEX}, inboth model-free and model-based manners, which can outperform baselines by astable margin in various MuJoCo environments with sparse rewards. Compared withexisting sample-efficient online RL algorithms with general functionapproximations, \texttt{MEX} achieves similar sample efficiency while enjoyinga lower computational cost and is more compatible with modern deep RL methods.

Maximize to Explore: One Objective Function Fusing Estimation, Planning, and Exploration