Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Mean-Field Games With Finitely Many Players: Independent Learning and Subjectivity

Authors: Bora Yongacoglu, Gürdal Arslan, Serdar Yüksel

JMLR 2024 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We ran Algorithm 3 on a 20-player mean-field game with compressed observability, described below in (12). This game can be interpreted as a model for decision-making during an epidemic or as a model for vehicle use decisions in a traffic network. ... Using the game in (12), we ran 250 independent trials of self-play under Algorithm 3, were each trial consisted of 20 exploration phases and each exploration phase consisted for 25,000 stage games. ... Our results are summarized in Figures 2 and 3 and in Table 1. In Figure 2, we plot the frequency of subjective ϵ-equilibrium against the exploration phase index. ... Table 1: Frequency of Subjective ϵ-equilibrium. πk,τ denotes the policy for EP k during the τ th trial.
Researcher Affiliation Academia Bora Yongacoglu EMAIL Department of Electrical and Computer Engineering University of Toronto; G urdal Arslan EMAIL Department of Electrical Engineering University of Hawaii at Manoa; Serdar Y uksel EMAIL Department of Mathematics and Statistics Queen s University
Pseudocode Yes Algorithm 1: Independent Learning of (Subjective) Value Functions; Algorithm 2: Subjective ϵ-satisficing Policy Revision (for player i N); Algorithm 3: Independent Learning
Open Source Code No The paper does not provide any explicit statement about releasing source code for the methodology described, nor does it include any links to code repositories.
Open Datasets No The paper describes a custom-designed simulation environment and its parameters in Section 6, 'Simulation Study', rather than using a pre-existing or publicly available dataset. There is no mention or reference to any publicly accessible dataset.
Dataset Splits No The paper describes running simulations, not experiments on a dataset with traditional splits. It mentions '250 independent trials of self-play' and 'each exploration phase consisted for 25,000 stage games' in the context of simulation runs, which is not equivalent to dataset splits for training, validation, and testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or cloud computing specifications) used to run the simulations described in the 'Simulation Study' section.
Software Dependencies No The paper does not provide specific software names with version numbers for implementation (e.g., Python, PyTorch, TensorFlow, specific solvers, etc.).
Experiment Setup Yes The game G used for our simulation is given by the following list: G = (N, X, Y, A, {ϕi}i N, Ploc, c, γ, ν0, ). ... N = {1, 2, . . . , 20} is a set of 20 players. ... X = {bad, medium, good}. ... A = {go, wait, heal}. ... The stage cost function c : X (X) A R is given by c(s, ν, a) := Rgo 1{a = go} + Rbad 1{s = bad} + Rheal 1{a = heal} for all (s, ν, a) X (X) A, where Rgo = 5 is a reward for undertaking one s usual business, Rbad = 10 is a penalty for being in bad condition, and Rheal = 3 is the cost of seeking healthcare. The discount factor γ = 0.8, and ν0 (X) is the product uniform distribution: xi 0 Unif(X) for each i N and the random variables {xi 0}i N are jointly independent. ... Our chosen parameters were ϵ = 5, di 1.5, and ei 0.25.