Learning Deviation Payoffs in Simulation-Based Games

Sokota, Samuel; Ho, Caleb; Wiedenbeck, Bryce

doi:10.1609/aaai.v33i01.33012173

Cited by 11 publications

(10 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There also has been prior work on regression for rolesymmetric games, for given role assignments (Wiedenbeck, Yang, and Wellman 2018;Sokota, Ho, and Wiedenbeck 2019). Duong et al (2009) and Fearnley et al (2015) studied algorithms for inducing structure of graphical games.…”

Section: Multiagent Simulation For Game Model Learningmentioning

confidence: 99%

Structure Learning for Approximate Solution of Many-Player Games

Wellman

2020

AAAI

View full text Add to dashboard Cite

Games with many players are difficult to solve or even specify without adopting structural assumptions that enable representation in compact form. Such structure is generally not given and will not hold exactly for particular games of interest. We introduce an iterative structure-learning approach to search for approximate solutions of many-player games, assuming only black-box simulation access to noisy payoff samples. Our first algorithm, K-Roles, exploits symmetry by learning a role assignment for players of the game through unsupervised learning (clustering) methods. Our second algorithm, G3L, seeks sparsity by greedy search over local interactions to learn a graphical game model. Both algorithms use supervised learning (regression) to fit payoff values to the learned structures, in compact representations that facilitate equilibrium calculation. We experimentally demonstrate the efficacy of both methods in reaching quality solutions and uncovering hidden structure, on both perfectly and approximately structured game instances.

show abstract

Section: Multiagent Simulation For Game Model Learningmentioning

confidence: 99%

Structure Learning for Approximate Solution of Many-Player Games

Wellman

2020

AAAI

View full text Add to dashboard Cite

show abstract

“…Given payoff data generated over a range of sampled profiles, induction of a game model can be viewed as a statistical machine learning problem. A variety of methods for learning game models from data have been developed (Honorio and Ortiz, 2015;Li and Wellman, 2020;Sokota et al, 2019;Vorobeychik et al, 2007;Wiedenbeck et al, 2018). Basing a game model on data-even simulated datamay overcome skepticism that some harbor about game-theoretic models.…”

Section: Incomplete Game Modelsmentioning

confidence: 99%

Economic Reasoning from Simulation-Based Game Models

Wellman¹

2020

oeconomia

View full text Add to dashboard Cite

Simulation modeling in economics has historically been viewed as an alternative to mainstream analytic technique, and as such has generally and intentionally avoided the focus on rational behavior and equilibrium reasoning that is characteristic of game-theoretic approaches. The emerging methodology of empirical game-theoretic analysis (EGTA) attempts to bridge agent-based simulation and game theory, combining the flexibility of simulation with the discipline of rationality criteria. The basic idea is to estimate a restricted game model from data generated by systematic simulation over a space of heuristic strategy profiles. EGTA enables a standard form of strategic analysis for complex economic scenarios previously addressed only in severely stylized form.

show abstract

“…A similar approach, which is still based on learning payoff functions using regression, is adopted by some recent works studying 1 The complete proofs of our theoretical results are available in Appendices A, B, and C. finite SBGs with many symmetric players [21,30]. Their goal is to exploit the symmetries so as to learn symmetric NEs in large games efficiently.…”

Section: Related Workmentioning

confidence: 99%

“…Wiedenbeck et al [30] focus on GP regression, since, as they show experimentally, it leads to better performances compared to other techniques. Subsequently, Sokota et al [21] provide an advancement over the previous work, using neural networks to approximate the utility function (instead of GPs) and providing a way to guide sampling so as to focus it on the neighborhood of candidate equilibrium points. These works significantly depart from ours, since (i) they aim at finding symmetric NEs in large SBGs with many symmetric players, (ii) they are restricted to games with finite strategy spaces, and (iii) they do not provide any theoretical guarantees on the quality of the obtained solutions.…”

Section: Related Workmentioning

confidence: 99%

“…Despite the modeling power of SBGs, recent works studying such games are only sporadic, addressing specific settings such as, e.g., symmetric games with a large number of players [21,30], empirical mechanism design [25], and two-player zero-sum finite games [9] (see Section 2 for more details and additional related works). To the best of our knowledge, the majority of these works focus on the case in which each player has a finite number of strategies available.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Learning Probably Approximately Correct Maximin Strategies in Simulation-Based Games with Infinite Strategy Spaces

Marchesi,

Trovò,

Gatti

2019

Preprint

View full text Add to dashboard Cite

We tackle the problem of learning equilibria in simulation-based games. In such games, the players' utility functions cannot be described analytically, as they are given through a black-box simulator that can be queried to obtain noisy estimates of the utilities. This is the case in many real-world games in which a complete description of the elements involved is not available upfront, such as complex military settings and online auctions. In these situations, one usually needs to run costly simulation processes to get an accurate estimate of the game outcome. As a result, solving these games begets the challenge of designing learning algorithms that can find (approximate) equilibria with high confidence, using as few simulator queries as possible. Moreover, since running the simulator during the game is unfeasible, the algorithms must first perform a pure exploration learning phase and, then, use the (approximate) equilibrium learned this way to play the game. In this work, we focus on two-player zero-sum games with infinite strategy spaces. Drawing from the best arm identification literature, we design two algorithms with theoretical guarantees to learn maximin strategies in these games. The first one works in the fixed-confidence setting, guaranteeing the desired confidence level while minimizing the number of queries. Instead, the second algorithm fits the fixedbudget setting, maximizing the confidence without exceeding the given maximum number of queries. First, we formally prove δ -PAC theoretical guarantees for our algorithms under some regularity assumptions, which are encoded by letting the utility functions be drawn from a Gaussian process. Then, we experimentally evaluate our techniques on a testbed made of randomly generated games and instances representing simple real-world security settings.

show abstract

Learning Deviation Payoffs in Simulation-Based Games

Cited by 11 publications

References 16 publications

Structure Learning for Approximate Solution of Many-Player Games

Structure Learning for Approximate Solution of Many-Player Games

Economic Reasoning from Simulation-Based Game Models

Learning Probably Approximately Correct Maximin Strategies in Simulation-Based Games with Infinite Strategy Spaces

Contact Info

Product

Resources

About