α-Rank: Multi-Agent Evaluation by Evolution.
Journal
Scientific reports
ISSN: 2045-2322
Titre abrégé: Sci Rep
Pays: England
ID NLM: 101563288
Informations de publication
Date de publication:
09 07 2019
09 07 2019
Historique:
received:
18
02
2019
accepted:
11
06
2019
entrez:
11
7
2019
pubmed:
11
7
2019
medline:
11
7
2019
Statut:
epublish
Résumé
We introduce α-Rank, a principled evolutionary dynamics methodology, for the evaluation and ranking of agents in large-scale multi-agent interactions, grounded in a novel dynamical game-theoretic solution concept called Markov-Conley chains (MCCs). The approach leverages continuous-time and discrete-time evolutionary dynamical systems applied to empirical games, and scales tractably in the number of agents, in the type of interactions (beyond dyadic), and the type of empirical games (symmetric and asymmetric). Current models are fundamentally limited in one or more of these dimensions, and are not guaranteed to converge to the desired game-theoretic solution concept (typically the Nash equilibrium). α-Rank automatically provides a ranking over the set of agents under evaluation and provides insights into their strengths, weaknesses, and long-term dynamics in terms of basins of attraction and sink components. This is a direct consequence of the correspondence we establish to the dynamical MCC solution concept when the underlying evolutionary model's ranking-intensity parameter, α, is chosen to be large, which exactly forms the basis of α-Rank. In contrast to the Nash equilibrium, which is a static solution concept based solely on fixed points, MCCs are a dynamical solution concept based on the Markov chain formalism, Conley's Fundamental Theorem of Dynamical Systems, and the core ingredients of dynamical systems: fixed points, recurrent sets, periodic orbits, and limit cycles. Our α-Rank method runs in polynomial time with respect to the total number of pure strategy profiles, whereas computing a Nash equilibrium for a general-sum game is known to be intractable. We introduce mathematical proofs that not only provide an overarching and unifying perspective of existing continuous- and discrete-time evolutionary evaluation models, but also reveal the formal underpinnings of the α-Rank methodology. We illustrate the method in canonical games and empirically validate it in several domains, including AlphaGo, AlphaZero, MuJoCo Soccer, and Poker.
Identifiants
pubmed: 31289288
doi: 10.1038/s41598-019-45619-9
pii: 10.1038/s41598-019-45619-9
pmc: PMC6617105
doi:
Types de publication
Journal Article
Research Support, Non-U.S. Gov't
Research Support, U.S. Gov't, Non-P.H.S.
Langues
eng
Sous-ensembles de citation
IM
Pagination
9937Références
Proc Natl Acad Sci U S A. 2013 Jan 22;110(4):1232-6
pubmed: 23297213
Nature. 2017 Oct 18;550(7676):354-359
pubmed: 29052630
Nature. 2016 Jan 28;529(7587):484-9
pubmed: 26819042
Phys Rev E Stat Nonlin Soft Matter Phys. 2006 Jul;74(1 Pt 1):011909
pubmed: 16907129
Phys Rev Lett. 2005 Dec 2;95(23):238701
pubmed: 16384353
Science. 2004 Feb 6;303(5659):793-9
pubmed: 14764867
Proc Natl Acad Sci U S A. 2002 Apr 2;99(7):4748-51
pubmed: 11930020
J Theor Biol. 2011 Apr 7;274(1):30-5
pubmed: 21232542
Phys Rev Lett. 2012 Apr 13;108(15):158104
pubmed: 22587290
Proc Natl Acad Sci U S A. 2017 Jul 3;114(27):E5396-E5405
pubmed: 28630336
Sci Rep. 2018 Jan 17;8(1):1015
pubmed: 29343692
J Math Biol. 1996;34(5-6):675-88
pubmed: 8691089
Phys Rev E Stat Nonlin Soft Matter Phys. 2006 Aug;74(2 Pt 1):021905
pubmed: 17025470
Science. 2018 Dec 7;362(6419):1140-1144
pubmed: 30523106
Science. 2017 May 5;356(6337):508-513
pubmed: 28254783