Evolutionary Dynamics and Phi-Regret Minimization in Games

Authors: Georgios Piliouras, Mark Rowland, Shayegan Omidshafiei, Romuald Elie, Daniel Hennes, Jerome Connor, Karl Tuyls

JAIR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We subsequently conduct experiments validating our theoretical results in a suite of 144 2 2 games wherein RD exhibits a diverse set of behaviors. We conclude by providing empirical evidence of Φ-regret minimization by RD in some larger games, hinting at further opportunity for Φ-regret based study of such algorithms from both a theoretical and empirical perspective.
Researcher Affiliation Collaboration Georgios Piliouras EMAIL Singapore University of Technology and Design Singapore Mark Rowland EMAIL Shayegan Omidshafiei EMAIL Romuald Elie EMAIL Daniel Hennes EMAIL Jerome Connor EMAIL Karl Tuyls EMAIL Deep Mind London, United Kingdom
Pseudocode No The paper does not contain any sections explicitly labeled as 'Pseudocode' or 'Algorithm'.
Open Source Code No The paper does not contain an explicit statement about releasing source code for the methodology described, nor does it provide a direct link to a code repository.
Open Datasets Yes The specific games we consider are those defined by Bruns (2015), which taxonomizes a collection of 2 2 games into several distinct classes by considering the patterns of payoffs received by each player. Bruns (2015) identifies 12 sets of basis payoffs corresponding to canonical games of varying characteristics, summarized for the row player in Fig. 5(a); corresponding column player payoffs for these games are defined in Bruns (2015) as the transpose along the anti-diagonal, which we also use for consistency. ... Each row of Fig. 8 visualizes results associated with a distinct 3 3 game, where the respective row player payoffs are defined as (12) with corresponding column player payoffs Bi = Ai for each game i.
Dataset Splits No For each of these games, we run 10 trials of RD (each with an independently-initialized set of seed strategies for the row and column players). The paper describes running multiple trials for each game but does not specify any training/test/validation dataset splits.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers used for the experiments.
Experiment Setup Yes For each of these games, we run 10 trials of RD (each with an independently-initialized set of seed strategies for the row and column players). ... The partitioning scheme Σ we use for computing mosaic regret is to subdivide the first player s mixed strategy space x into 10 discrete, equally-sized bins.