Learning Optimal Tax Design in Nonatomic Congestion Games
Authors: Qiwen Cui, Maryam Fazel, Simon S. Du
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We implemented our algorithm and conducted experiments on a classic example known as the nonlinear variant of Pigou s example [Nisan et al., 2007]. Concretely, nonlinear variant of Pigou s example is a routing game with one source node s and one target node t. There are two edges connecting s and t. One edge has constant cost c0(x) = c, x [0, 1] for some c [0, 1], and the other edge has polynomial cost c1(x) = xp. One important property of such games is the price of anarchy grows without bound as p , which urges proper tax to induce socially optimal behavior. We apply our algorithm to learn the optimal tax with different c0 and p. As we can see, the social welfare quickly converges to the optimal one. Another important observation is the learned tax function does not uniformly converge to the marginal cost tax, which is reasonable as accurate estimate is only necessary around the Nash equilibrium induced by the tax. |
| Researcher Affiliation | Academia | Qiwen Cui Paul G. Allen School of Computer Science Engineering University of Washington Seattle, WA 98195 qwcui@cs.washington.edu Maryam Fazel Department of Electrical Computer Engineering University of Washington Seattle, WA 98195 mfazel@uw.edu Simon S. Du Paul G. Allen School of Computer Science Engineering University of Washington Seattle, WA 98195 ssdu@cs.washington.edu |
| Pseudocode | Yes | Algorithm 1 Tax Design for Congestion Game Algorithm 2 Test Tax Design |
| Open Source Code | No | The NeurIPS checklist for this paper explicitly states 'No' for open access to code, and no code links or statements are found in the paper's main content or appendices. |
| Open Datasets | No | The paper uses a 'nonlinear variant of Pigou’s example' which describes a theoretical game setup rather than a pre-existing publicly available dataset. It does not provide any specific links, DOIs, or citations for a dataset used. |
| Dataset Splits | No | The paper does not explicitly provide training, validation, or test dataset splits. The experiment is conducted on a simulated game example rather than a dataset with predefined splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software dependencies with version numbers (e.g., Python, PyTorch, or specific solvers with versions) needed to replicate the experiment. |
| Experiment Setup | No | The paper describes the specific game example parameters (c and p for the cost functions) but does not provide typical experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs) or optimizer settings for the proposed algorithm. |