Kagan Tumer's Publications

Display Publications by [Year] [Type] [Topic]


Using Reward/Utility Based Impact Scores in Partitioning (Extended Abstract). W. Curran, A. Agogino, and K. Tumer. In Proceedings of the Thirteenth International Joint Conference on Autonomous Agents and Multiagent Systems, pp. , Paris, France, May 2014.

Abstract

Reinforcement learning with reward shaping is a natural approach to solving large multiagent domains with agents who must cooperate to achieve some system objective. However, reward shaping can be computationally expensive to compute. Agent partitioning can assist in this computational complexity by treating each partition of agents as an independent problem. In this paper we introduce a novel agent partitioning approach called Reward/Utility-Based Impact (RUBI). RUBI finds an effective partitioning of agents while requiring no prior knowledge about the domain, leads to better performance by discovering a non-trivial agent partitioning, and leads to faster simulations. We test RUBI in the Air Traffic Flow Management Problem (ATFMP), where there are simultaneously tens of thousands of aircraft affecting the system and no intuitively accurate similarity metric between agents. When partitioning with RUBI in the ATFMP, there is a 37\% increase in performance, with a 510x speed up per simulation step over non-partitioning approaches. Additionally, RUBI matches the performance of the current ATFMP gold standard using no prior knowledge and performing each learning step 10\% faster.

Download

[PDF]202.3kB  

BibTeX Entry

@inproceedings{tumer-curran_aamas14,
        author = {W. Curran and A. Agogino and  K. Tumer},
        title = {Using Reward/Utility Based Impact Scores in Partitioning (Extended Abstract)},
        booktitle = {Proceedings of the Thirteenth International Joint Conference on Autonomous Agents and Multiagent Systems},
	month = {May},
          pages ={},
	address = {Paris, France},
	abstract={Reinforcement learning with reward shaping is a natural approach to solving large multiagent domains with agents who must cooperate to achieve some system objective. However, reward shaping can be computationally expensive to compute. Agent partitioning can assist in this computational complexity by treating each partition of agents as an independent problem. In this paper we introduce a novel agent partitioning approach called Reward/Utility-Based Impact (RUBI). RUBI finds an effective partitioning of agents while requiring no prior knowledge about the domain, leads to better performance by discovering a non-trivial agent partitioning, and leads to faster simulations. We test RUBI in the Air Traffic Flow Management Problem (ATFMP), where there are simultaneously tens of thousands of aircraft affecting the system and no intuitively accurate similarity metric between agents. When partitioning with RUBI in the ATFMP, there is a 37\% increase in performance, with a 510x speed up per simulation step over non-partitioning approaches. Additionally, RUBI matches the performance of the current ATFMP gold standard using no prior knowledge and performing each learning step 10\% faster.},
	bib2html_pubtype = {Refereed Conference Papers},
	bib2html_rescat = {Reinforcement Learning, Multiagent Systems},
        year = {2014}
}

Generated by bib2html.pl (written by Patrick Riley ) on Tue Jun 26, 2018 19:10:42