Learning Deep Mean Field Games for Modeling Large Population Behavior

Authors: Jiachen Yang, Xiaojing Ye, Rakshit Trivedi, Huan Xu, Hongyuan Zha

ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our method learns both the reward function and forward dynamics of an MFG from real data, and we report the first empirical test of a mean field game model of a real-world social media population.
Researcher Affiliation Academia 1Georgia Institute of Technology 2Georgia State University
Pseudocode Yes Algorithm 1 Guided cost learning; Algorithm 2 Actor-critic algorithm for MFG
Open Source Code No The paper does not include a statement about releasing source code or a link to a code repository.
Open Datasets No We use data representing the activity of a Twitter population consisting of 406 users. Data collection proceeded as follows for 27 days...
Dataset Splits No The paper mentions 'The training set consists of trajectories...over the first M = 21 days' and 'evaluated against data from 6 held-out test days', but does not explicitly detail a separate validation set or its split.
Hardware Specification No The paper does not specify any hardware details used for running the experiments.
Software Dependencies No VAR was implemented using the Statsmodels module in Python, with order 18 selected via random sub-sampling validation with validation set size 5 (Seabold & Perktold, 2010). All layers were initialized using the Xavier normal initializer in Tensorflow.
Experiment Setup Yes Table 1: Parameters S max actor-critic episodes 4000 β critic learning rate O(1/s) ξ actor learning rate O(1/s ln ln s) c αi j scaling factor 1e4 ϵ Adam optimizer learning rate for reward 1e-4 d R convergence threshold for reward iteration 1e-4 θfinal learned policy parameter 8.64