Anticipatory Fictitious Play

Authors: Alex Cloud, Albert Wang, Wesley Kerr

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct an extensive comparison of our algorithm with fictitious play, proving an optimal O(t 1) convergence rate for certain classes of games, demonstrating superior performance numerically across a variety of games, and concluding with experiments that extend these algorithms to the setting of deep multiagent reinforcement learning.
Researcher Affiliation Industry Alex Cloud , Albert Wang and Wesley Kerr Riot Games kacloud@gmail.com, {alwang, wkerr}@riotgames.com
Pseudocode Yes Algorithm 1 Neu PL-FP/AFP
Open Source Code No The paper does not include an unambiguous statement that the authors are releasing code for the work described, nor does it provide a direct link to such source code.
Open Datasets Yes We use two environments, our own Tiny Fighter, and Running With Scissors, from [Vezhnevets et al., 2020]. [Vezhnevets et al., 2020] and [Liu et al., 2022] (Appendix B.1.) feature further discussion of the environment.
Dataset Splits No The paper describes the training process and evaluation metrics but does not provide specific details on dataset split percentages or counts for training, validation, and testing.
Hardware Specification No For reinforcement learning, we use the Asynchronous Proximal Policy Optimization (APPO) algorithm [Schulman et al., 2017], a distributed actor-critic RL algorithm, as implemented in RLLib [Moritz et al., 2018] with a single GPU learner.
Software Dependencies No For reinforcement learning, we use the Asynchronous Proximal Policy Optimization (APPO) algorithm [Schulman et al., 2017], a distributed actor-critic RL algorithm, as implemented in RLLib [Moritz et al., 2018] with a single GPU learner.
Experiment Setup No Hyperparameter settings are given in the supplementary material, Section E.