Display Publications by [Year] [Type] [Topic]
Learning correct behavior from one example (one- shot learning) is particularly difficult in multiagent systems where the pertinent information is potentially distributed across agents, and the emergent behavior of the system is dependent on inter-agent interactions. This paper introduces Distributed Modular Memory Units (DMMU), a distributed multiagent learning framework. DMMU uses a shared external memory to enable one-shot adaptive learning in multiagent systems. The external memory is accessed by agents acting independently and in parallel. Agents operate by processing their individual states and deciding when to interact with the shared external memory unit. This allows agents to identify, retain, and propagate information among the team. Subsequently, DMMU can rapidly assimilate task features from a group of distributed agents and successfully use the information to accomplish distributed one- shot learning. We compare the performance of the DMMU framework on a simulated cybersecurity task with traditional feed-forward ensembles, long short-term memory based agents, and a centralized learner. Results demonstrate that DMMU shows at least 100% performance improvement over the other methods, and exhibits distributed one-shot learning to effectively solve this complex task.
(unavailable)
@InProceedings{tumer-khadka_mrs19,
author = {S. Khadka and C. Yates and K. Tumer},
title = {Memory-based Multiagent One-Shot Learning (Poster session)},
booktitle = {IEEE International Symposium on Multi-Robot and Multi-Agent Systems (MRS)},
address = {Rutgers, NJ},
month = {August},
pages={},
abstract={Learning correct behavior from one example (one- shot learning) is particularly difficult in multiagent systems where the pertinent information is potentially distributed across agents, and the emergent behavior of the system is dependent on inter-agent interactions. This paper introduces Distributed Modular Memory Units (DMMU), a distributed multiagent learning framework. DMMU uses a shared external memory to enable one-shot adaptive learning in multiagent systems. The external memory is accessed by agents acting independently and in parallel. Agents operate by processing their individual states and deciding when to interact with the shared external memory unit. This allows agents to identify, retain, and propagate information among the team. Subsequently, DMMU can rapidly assimilate task features from a group of distributed agents and successfully use the information to accomplish distributed one- shot learning. We compare the performance of the DMMU framework on a simulated cybersecurity task with traditional feed-forward ensembles, long short-term memory based agents, and a centralized learner. Results demonstrate that DMMU shows at least 100% performance improvement over the other methods, and exhibits distributed one-shot learning to effectively solve this complex task.},
bib2html_pubtype = {Refereed Conference Papers},
bib2html_rescat = {Multiagent Systems},
year = {2019}
}
Generated by bib2html.pl (written by Patrick Riley ) on Wed Apr 01, 2020 17:39:43