Near-Optimal No-Regret Learning Dynamics for General Convex Games

Authors: Gabriele Farina, Ioannis Anagnostides, Haipeng Luo, Chung-Wei Lee, Christian Kroer, Tuomas Sandholm

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments Finally, while our main contribution is of theoretical nature, we also support our theory by conducting experiments on some standard extensive-form games (Appendix C). The experiments verify that under LRL-OFTRL the regret of each player grows as O(log T).
Researcher Affiliation Collaboration Gabriele Farina Carnegie Mellon University Pittsburgh, PA 15213 gfarina@cs.cmu.edu Ioannis Anagnostides Carnegie Mellon University Pittsburgh, PA 15213 ianagnos@cs.cmu.edu Haipeng Luo University of Southern California Los Angeles, CA 90007 haipengl@usc.edu Chung-Wei Lee University of Southern California Los Angeles, CA 90007 leechung@usc.edu Christian Kroer Columbia University New York, NY 10027 christian.kroer@columbia.edu Tuomas Sandholm Carnegie Mellon University Strategy Robot, Inc. Optimized Markets, Inc. Strategic Machine, Inc. Pittsburgh, PA 15213 sandholm@cs.cmu.edu
Pseudocode Yes Algorithm 1: Log-Regularized Lifted Optimistic FTRL (LRL-OFTRL)
Open Source Code Yes The code to reproduce our experiments is publicly available at: https://github.com/anonymous_author/LRL-OFTRL
Open Datasets No The paper refers to standard game types (e.g., Extensive-Form Games, Goofspiel, Leduc Poker) as the environment for experiments. While these games are well-defined, the paper does not refer to them as 'datasets' in the traditional sense, nor does it provide links/citations for data splits or access to specific data files typically associated with public datasets for training machine learning models.
Dataset Splits No The paper conducts experiments in games but does not specify training, validation, or test dataset splits in the conventional machine learning sense, as it focuses on online learning dynamics in game theory.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify versions for any software dependencies, libraries, or programming languages used in the implementation.
Experiment Setup Yes If all players follow LRL-OFTRL with learning rate η min n 1 256B X 1 , 1 128n L X 2 1 o... For each game, we ran the algorithm for T = 10^5 iterations using the optimal learning rate η = 1/2 for Goofspiel and η = 1/4 for Leduc Poker.