Many multi-agent reinforcement learning (MARL) problems are well-suited to be solved in a distributed manner. Not only to achieve scalability in terms of runtime or memory, but also because the environment may be a distributed system. While distributed machine learning is appealing in these cases, it does introduce several complexities inherent to distributed computing This paper presents Marlon, a domain-specific language to facilitate MARL in a distributed environment. Marlon essentially acts as a bridge between the environment and the agents observing it, and abstracts away many of the concerns related to distributed computing. We demonstrate Marlon by means of a load balancing example, where MARL is used to optimize the system’s performance, while allowing network nodes to join and leave the system.
Original languageEnglish
Title of host publicationBenelux Conference on Artificial Intelligence
PublisherBNAIC 2018, Belgium-Netherlands conference on Artificial Intelligence
Pages169-170
Number of pages2
Volume1
Publication statusPublished - 2018
Event30th Benelux Conference on Artificial Intelligence - BNAIC 2018 - Jheronimus Academy of Data Science, ‘s-Hertogenbosch, Netherlands
Duration: 8 Nov 20189 Nov 2018
https://bnaic2018.nl

Conference

Conference30th Benelux Conference on Artificial Intelligence - BNAIC 2018
CountryNetherlands
City‘s-Hertogenbosch
Period8/11/189/11/18
Internet address

    Research areas

  • Domain-Specific Languages, Multi-agent Reinforcement Learning, Distributed Systems

ID: 40239092