Accelerated Regularized Learning in Finite N-Person Games

Authors: Kyriakos Lotidis, Angeliki Giannou, Panayotis Mertikopoulos, Nicholas Bambos

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conclude this section with a series of numerical simulations to validate the performance of (FTXL).
Researcher Affiliation Academia Kyriakos Lotidis Stanford University klotidis@stanford.edu; Angeliki Giannou University of Wisconsin Madison giannou@wisc.edu; Panayotis Mertikopoulos Univ. Grenoble Alpes, CNRS, Inria, Grenoble INP LIG 38000 Grenoble, France panayotis.mertikopoulos@imag.fr; Nicholas Bambos Stanford University bambos@stanford.edu
Pseudocode Yes 𝑦𝑖,𝑛+1 = 𝑦𝑖,𝑛+ 𝛾𝑝𝑖,𝑛+1; 𝑝𝑖,𝑛+1 = 𝑝𝑖,𝑛+ 𝛾ˆ𝑣𝑖,𝑛; π‘₯𝑖,𝑛= 𝑄𝑖(𝑦𝑖,𝑛) . (FTXL)
Open Source Code Yes Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The code is included in the supplemental material.
Open Datasets No The paper does not use a publicly available or open dataset in the traditional sense for training. Instead, it defines the parameters of specific game types (zero-sum game, congestion game) for its numerical simulations. No concrete access information like a link, DOI, or citation to a pre-existing dataset is provided.
Dataset Splits No The paper defines specific game parameters for its simulations rather than using external datasets with explicit train/validation/test splits. It conducts '100 separate trials' but does not define a separate validation split within these trials.
Hardware Specification Yes The experiments have been implemented using Python 3.11.5 on a M1 Mac Book Air with 16GB of RAM.
Software Dependencies Yes The experiments have been implemented using Python 3.11.5 on a M1 Mac Book Air with 16GB of RAM.
Experiment Setup Yes For the zero-sum game, all runs were initialized with 𝑦1 = 0, and we used constant step-size 𝛾= 10 2, and exploration parameter πœ€= 10 1, where applicable. For the congestion game, the initial state 𝑦1 for each run was drawn uniformly at random in [ 1, 1]2, and we used constant step-size 𝛾= 10 2, and exploration parameter πœ€π‘›= 1/𝑛1/4, where applicable.