A Simple, Fast, and Safe Mediator for Congestion Management
Authors: Kei Ikegami, Kyohei Okumura, Takumi Yoshikawa2030-2037
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our first set of experiments shows that the mediator works well in more general settings. Even when there are more than two resources and agents have heterogeneous opportunity costs, the approximation rate of ex-post Nash equilibrium is bounded by the value inferred from the theoretical results on simple cases. Our experiments also confirm that our mediator does satisfy the safe participation property in the general cases. |
| Researcher Affiliation | Academia | Kei Ikegami,1 Kyohei Okumura,1 Takumi Yoshikawa1 1Graduate School of Economics, The University of Tokyo {ikegami-kei0120, k-okumura, takumi-yoshikawa157}@g.ecc.u-tokyo.ac.jp |
| Pseudocode | Yes | Algorithm 1 ϵ-BRD(τ = (τ1, , τm) , ϵ) |
| Open Source Code | No | The paper does not provide a direct link to open-source code for the described methodology nor an explicit statement about its public release. It mentions plans for a web application in the future but not current code availability. |
| Open Datasets | No | The paper describes simulations and experiments in congestion games with varying parameters (e.g., number of agents, resources) but does not use or provide access to any specific public dataset. The data used for experiments appears to be generated based on the defined game parameters rather than being a pre-existing public dataset. |
| Dataset Splits | No | The paper discusses simulation setups and parameter variations but does not describe training, validation, or test dataset splits in the context of typical machine learning experiments. The experiments are simulations of game theory dynamics. |
| Hardware Specification | No | The paper describes simulations and experimental results but does not specify any hardware details (e.g., GPU/CPU models, memory) used to conduct these experiments. |
| Software Dependencies | No | The paper refers to its proposed mediator (CRAB) and Best Response Dynamics (BRD) but does not provide specific names and version numbers for any software dependencies, libraries, or programming languages used for its implementation or experiments. |
| Experiment Setup | Yes | The experimental details, including the precise parameter settings, are provided in the Appendix. We conduct experiments in a two-resource setting. We try 2, 160 parameters made from reasonable values. We observe that our specification of BRD outperforms other versions both in the realized social welfare and in the computation time. In order to define BRD precisely, we need to fix the choice rule and the initialization rule, neither of which the usual definition of BRD precisely determines. The choice rule specifies who moves at each step of BRD. Here, we consider two options: (1) to pick the agent with the maximum payoff gain (max increment), or (2) to pick the one with the minimum payoff gain (min increment). As for the initialization rule, we compare three alternatives: initially, all agents choose (1) their best preferred resources (best), (2) their worst preferred ones (worst), and (3) the same one resource (oneside). |