Subgame Solving in Adversarial Team Games

Authors: Brian Zhang, Luca Carminati, Federico Cacciamani, Gabriele Farina, Pierriccardo Olivieri, Nicola Gatti, Tuomas Sandholm

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our method to a standard test suite, and we empirically show the performance improvement of the strategies thanks to subgame solving.
Researcher Affiliation Collaboration Brian Hu Zhang Computer Science Department Carnegie Mellon University bhzhang@cs.cmu.edu Luca Carminati DEIB, Politecnico di Milano luca.carminat@polimi.it Federico Cacciamani DEIB, Politecnico di Milano federico.cacciamani@polimi.it Gabriele Farina Computer Science Department Carnegie Mellon University gfarina@cs.cmu.edu Pierriccardo Olivieri DEIB, Politecnico di Milano pierriccardo.oliveri@polimi.it Nicola Gatti DEIB, Politecnico di Milano nicola.gatti@polimi.it Tuomas Sandholm Computer Science Department, CMU Strategic Machine, Inc. Strategy Robot, Inc. Optimized Markets, Inc. sandholm@cs.cmu.edu
Pseudocode Yes Algorithm 1 Maxmargin subgame solving with column generation, at public state P
Open Source Code No The paper states in its self-assessment checklist that code is included, but the provided text does not contain a direct link to a code repository, nor an explicit statement in the main body or appendices (other than results in Appendix C) indicating where the code for the described methodology can be accessed.
Open Datasets No The paper uses "parametric versions of the ATG instances customarily adopted in the literature" such as Kuhn poker, Leduc poker, Liar's Dice, and Tricks, citing papers [11, 18, 13, 21] for their rules. These are game definitions/frameworks, not publicly available datasets in the traditional sense with specific access information (link, DOI, repository) for data files.
Dataset Splits No The paper does not specify traditional training/validation/test dataset splits. It describes using game instances and parameters for those games, rather than data splits.
Hardware Specification Yes Each experiment was allocated 32 CPU cores and 256 GB RAM on a cluster machine.
Software Dependencies Yes Integer and linear programs were solved with Gurobi 9.5.
Experiment Setup Yes More precisely, the blueprint computation is stopped once 10 minutes have elapsed or column generation has achieved a Nash gap of /10, where is the difference between maximum and minimum team s payoffs, whichever comes first. We use a range of time limits for the strategy refinement, defined as the average time needed by a single iteration of the CG algorithm at the root of the whole game multiplied by a number α {0, 1, ..., 10}.