Worst-Case VCG Redistribution Mechanism Design Based on the Lottery Ticket Hypothesis
Authors: Mingyu Guo
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The results are summarized in Table 1. The quality of training is naturally the gap between the achieved worst-case allocative efficiency ratio and the theoretical upper bound. We use WCT to denote our worst-case training algorithm. |
| Researcher Affiliation | Academia | School of Computer and Mathematical Sciences University of Adelaide, Australia mingyu.guo@adelaide.edu.au |
| Pseudocode | Yes | Algorithm 1: Worst-Case Training Algorithm |
| Open Source Code | No | The paper does not provide any statement about releasing source code or a link to a code repository for the described methodology. |
| Open Datasets | No | The paper does not use pre-existing public datasets; instead, it generates 'random type profiles' and 'worst-case type profiles' as training samples. No links or citations for publicly available datasets are provided. |
| Dataset Splits | No | The paper describes how training batches are composed of generated type profiles but does not specify train/validation/test splits of a fixed dataset or reference standard split methodologies for reproducibility. |
| Hardware Specification | Yes | The hardware allocated to each job is 1 CPU core from Intel Xeon Platinum 8360Y (for running MIPs) and 1 GPU core from Nvidia A100 (for neural network training). |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., specific libraries, frameworks, or solvers with their versions). |
| Experiment Setup | Yes | Run Adam SGD on h with learning rate 0.0001 for 500 epochs. Training batch consists of: 16 latest calculated worst-case type profiles (i.e., WCP[-16:]) 16 randomly sampled worst-case type profiles from earlier (i.e., from WCP[:-16]) 16 random type profiles n + 1 type profiles where the agents either report 1 n/2 or 0 (i.e., type profiles for deriving the conjectured upper bound (Naroditskiy et al. 2012)) |