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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Decentralized execution is one core demand in cooperative multi-agentreinforcement learning (MARL). Recently, most popular MARL algorithms haveadopted decentralized policies to enable decentralized execution and usegradient descent as their optimizer. However, there is hardly any theoreticalanalysis of these algorithms taking the optimization method into consideration,and we find that various popular MARL algorithms with decentralized policiesare suboptimal in toy tasks when gradient descent is chosen as theiroptimization method. In this paper, we theoretically analyze two common classesof algorithms with decentralized policies -- multi-agent policy gradientmethods and value-decomposition methods to prove their suboptimality whengradient descent is used. In addition, we propose the Transformation AndDistillation (TAD) framework, which reformulates a multi-agent MDP as a specialsingle-agent MDP with a sequential structure and enables decentralizedexecution by distilling the learned policy on the derived ``single-agent" MDP.This approach uses a two-stage learning paradigm to address the optimizationproblem in cooperative MARL, maintaining its performance guarantee.Empirically, we implement TAD-PPO based on PPO, which can theoretically performoptimal policy learning in the finite multi-agent MDPs and shows significantoutperformance on a large set of cooperative multi-agent tasks.

Towards Global Optimality in Cooperative MARL with the Transformation And Distillation Framework