Learning with Linear Function Approximations in Mean-Field Control

Authors: Erhan Bayraktar, Ali Devran Kara

JMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Numerical Study We now present a numerical example to verify the results we have established in the earlier sections. We consider a multi-agent taxi service model where each agent represents a taxi. The state and action spaces are binary such that X = U = {0, 1}. [...] Figure 1 shows the value loss for different values of the number of agents in the system. We graph the loss functions under 3 settings:
Researcher Affiliation Academia Erhan Bayraktar EMAIL Department of Mathematics, University of Michigan, Ann Arbor, MI, USA Ali Devran Kara EMAIL Department of Mathematics, Florida State University, FL, USA
Pseudocode No The paper describes iterative learning algorithms using mathematical equations (e.g., equations (22), (23), (24), (25)) within the main text, but it does not feature any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper includes a license for the paper itself (CC-BY 4.0) but does not contain any explicit statements about releasing source code for the described methodology, nor does it provide links to code repositories.
Open Datasets No The paper describes a 'multi-agent taxi service model' with defined cost structures and dynamics for its numerical study. This appears to be a simulated environment or a model generated by the authors, and no external or publicly available dataset is referenced or provided with access information.
Dataset Splits No The paper presents a numerical study based on a simulated multi-agent taxi service model with defined parameters. Since it does not use any external dataset, there is no information about dataset splits (e.g., training, validation, test).
Hardware Specification No The paper presents a 'Numerical Study' section to verify its results but does not provide any specific details regarding the hardware (e.g., GPU or CPU models, memory specifications) used to run these numerical experiments.
Software Dependencies No The paper discusses various learning algorithms and a numerical study but does not explicitly mention any specific software libraries, frameworks, or their version numbers (e.g., Python, PyTorch, TensorFlow) that were used for implementation.
Experiment Setup Yes We set the parameters as R = 1, S = 7 and β = 0.7. [...] The estimate policy for the infinite population model, where the transition-cost function is learned using discretization basis functions based on the discretization of the measure space P(X) into 6 subsets (see Section 3.3). The estimate policy for the infinite population model, where the transition-cost function is learned using a class of basis functions: φ(µ) = [1, µ(0), µ(0)2, µ(0)3, sin(µ(0)), cos(µ(0))].