Large Scale Learning of Agent Rationality in Two-Player Zero-Sum Games

Authors: Chun Kai Ling, Fei Fang, J. Zico Kolter6104-6111

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental When tested on randomly generated games, we report speedups of orders of magnitude over previous approaches. We also demonstrate the effectiveness of our model on both real-world one-player settings and synthetic data. 5 Experiments The proposed first order method was implemented using Cython. ... Here we use randomly generated extensive form games to illustrate the computational efficiency of our proposed first order method compared to the second order method used by Ling, Fang, and Kolter.
Researcher Affiliation Collaboration Chun Kai Ling,1 Fei Fang,1 J. Zico Kolter1,2 1 School of Computer Science, Carnegie Mellon University 2 Bosch Center for Artificial Intelligence {chunkail, feif, zkolter}@cs.cmu.edu
Pseudocode Yes Algorithm 1: Learning game parameters Φ using SGD; Algorithm 2: FOM method to solve (9)
Open Source Code No The paper does not provide explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Here we demonstrate the applicability of the nested logit model (i.e., a one player game) using a publically available dataset (Hunt et al. 2016).
Dataset Splits Yes The training set of size 2000, with an independent test set of size 1000.
Hardware Specification No Experiments are run on the cloud with identical Amazon EC2 instances.
Software Dependencies No The paper mentions software like Cython, Numpy, Scipy, and PyTorch, but does not provide specific version numbers for these dependencies.
Experiment Setup Yes We minimized the log loss using the Adam optimizer with a batch size of 64 and learning rate of 10 4. We set τ = 1 for the forward solver and τ = 0.1 for the backward solver.