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

The goal of reinforcement learning (RL) is to maximize the expectedcumulative return. It has been shown that this objective can be represented byan optimization problem of the state-action visitation distribution underlinear constraints. The dual problem of this formulation, which we refer to asdual RL, is unconstrained and easier to optimize. We show that severalstate-of-the-art off-policy deep reinforcement learning (RL) algorithms, underboth online and offline, RL and imitation learning (IL) settings, can be viewedas dual RL approaches in a unified framework. This unification provides acommon ground to study and identify the components that contribute to thesuccess of these methods and also reveals the common shortcomings acrossmethods with new insights for improvement. Our analysis shows that prioroff-policy imitation learning methods are based on an unrealistic coverageassumption and are minimizing a particular f-divergence between the visitationdistributions of the learned policy and the expert policy. We propose a newmethod using a simple modification to the dual RL framework that allows forperformant imitation learning with arbitrary off-policy data to obtainnear-expert performance, without learning a discriminator. Further, by framinga recent SOTA offline RL method XQL in the dual RL framework, we proposealternative choices to replace the Gumbel regression loss, which achieveimproved performance and resolve the training instability issue of XQL. Projectcode and details can be found at this https://hari-sikchi.github.io/dual-rl.

Dual RL: Unification and New Methods for Reinforcement and Imitation Learning