Last-iterate Convergence in Regularized Graphon Mean Field Game

Authors: Jing Dong, Baoxiang Wang, Yaoliang Yu

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we verify the performance of the studied algorithms by empirically testing them against fictitious play in a variety of tasks. ... We validate the effectiveness of the studied algorithm by empirically comparing them against the fictitious play in four different environments. ... Figure 1: Experimental results for the mean field games described.
Researcher Affiliation Academia Jing Dong1,*, Baoxiang Wang1,3, , Yaoliang Yu2,3, 1 Chinese University of Hong Kong, Shenzhen, 2 University of Waterloo, 3 Vector Institute EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Tabular online mirror descent for λ-regularized GMFG
Open Source Code No No explicit statement or link for the authors' own code is provided. The mentions of 'Open Spiel' refer to a third-party implementation used for environments, not the authors' source code. For example: 'We use the setup of fictitious play from (Perrin et al. 2020) and follow the implementation from Open Spiel.'
Open Datasets Yes For all the environments described below, we use the Open Spiel implementation of the games. Predator Prey, Crowd Avoidance, Crowd modeling, and Periodic Aversion. ... Crowd Modeling ... (Perrin et al. 2020) ... Predator Prey ... (P erolat et al. 2022) ... Periodic Aversion ... (Almulla, Ferreira, and Gomes 2017)
Dataset Splits No The paper evaluates algorithms in simulated game environments (Predator Prey, Crowd Avoidance, Crowd modeling, Periodic Aversion) which are not typically partitioned into training, validation, and test sets in the traditional supervised learning sense. Therefore, no specific dataset split information is provided.
Hardware Specification No We ran all experiments with a 10-core CPU, with 32 GB memory. This mentions core count and memory amount but lacks a specific CPU model or type, falling short of the detailed hardware specification requirement.
Software Dependencies No The paper mentions using 'Open Spiel' for the implementation of the game environments but does not provide specific version numbers for Open Spiel or any other software libraries or dependencies.
Experiment Setup Yes For all of our experiments, we choose the learning rate to be ηt = 0.1 and the exploration rate γt = 0.1. We repeat the experiments with 5 different random seeds.