Team-Fictitious Play for Reaching Team-Nash Equilibrium in Multi-team Games
Authors: Ahmed Dönmez, Yüksel Arslantaş, Muhammed Sayin
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide extensive simulations of Team-FP dynamics and compare its performance with other widely studied dynamics such as smooth fictitious play and multiplicative weights update. We further explore how different parameters impact the speed of convergence. In this section, we present various simulation results demonstrating the coordination speed of Team FP and compare it to pure FP, no-regret algorithms, and a stationary opponent. |
| Researcher Affiliation | Academia | Ahmed Said Donmez Bilkent University said.donmez@bilkent.edu.tr Yuksel Arslantas Bilkent University yuksel.arslantas@bilkent.edu.tr Muhammed O. Sayin Bilkent University sayin@ee.bilkent.edu.tr |
| Pseudocode | Yes | Algorithm 1 (Independent) Team-FP; Algorithm 2 Model-based (Independent) Team-FP for MGs; Algorithm 3 Model-free (Independent) Team-FP for MGs |
| Open Source Code | Yes | We share code for experiments. However, the code may be too complex to understand as most of the documentation for the code is missing. |
| Open Datasets | No | No concrete access information (link, DOI, repository, or formal citation to a public dataset) is provided. The experiments use randomly generated game instances. |
| Dataset Splits | No | The paper does not describe specific training, validation, and test dataset splits as it focuses on game theory simulations rather than traditional machine learning tasks with pre-defined datasets. |
| Hardware Specification | Yes | All the simulations are executed on a computer equipped with an Intel Xeon W7-3455 CPU and 128 GB RAM. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., library names like PyTorch, TensorFlow, or specific solver versions) are provided. |
| Experiment Setup | Yes | For all simulations, temperature parameter τ is chosen to be 0.1 unless another option is mentioned. We conduct simulations for ZSPTG with two different setups: one with three teams, each consisting of three agents, and another with two teams, each consisting of four agents. Unless explicitly stated otherwise, the default setting consists of two teams, with four agents in each team. The step size is chosen to be αk = 1/(k + 1) for all simulations. |