Evolution Strategies for Approximate Solution of Bayesian Games

Authors: Zun Li, Michael P. Wellman5531-5540

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our methods on MBS and HG environments of 5, 10, and 15 agents with the same number of goods. Each experiment runs for 5 trials. The results are shown in Figures 2 and 3.
Researcher Affiliation Academia Zun Li and Michael P. Wellman University of Michigan, Ann Arbor {lizun, wellman}@umich.edu
Pseudocode Yes Algorithm 1: Natural Evolution Strategies; Algorithm 2: Minimax-NES for PBNE; Algorithm 3: Incremental Strategy Generation
Open Source Code Yes Li, Z.; and Wellman, M. P. 2021. Evolution strategies for approximate solution of Bayesian games: Supplementary material. Avaliable at https://rezunli96.github.io/.
Open Datasets No The paper describes the setup of simulated environments (MBS and HG) where types are drawn from distributions, but it does not refer to a pre-existing publicly available dataset with concrete access information (link, DOI, or formal citation).
Dataset Splits No The paper does not explicitly provide training, validation, and test splits with specific percentages, absolute sample counts, or references to predefined splits for the main experimental evaluation.
Hardware Specification No The paper does not explicitly describe the hardware used for running its experiments, such as specific GPU/CPU models, processors, or memory amounts. It only discusses the computational methods.
Software Dependencies No The paper mentions software components like 'Adam optimizer' and 'two-layer perceptron' but does not specify their version numbers or the versions of any underlying programming languages or libraries used for implementation.
Experiment Setup Yes The hyperparameters for NES are tuned as follows. We fixed population size J = 4 + 3 log d as the default setting adopted by Wierstra et al. (2014), where d is the number of parameters of the deep model. For every fitness function we apply a grid search to select the best bandwidth ν and learning rate α within certain ranges, to maximize the performance of the resulted NES. ... For black-box functions O( , s) and O( , σ), each query we run an agent-based simulation for 5000 times and take the corresponding average payoff values as the outputs.