Generalized Demographic Parity for Group Fairness
Authors: Zhimeng Jiang, Xiaotian Han, Chao Fan, Fan Yang, Ali Mostafavi, Xia Hu
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show the better bias mitigation performance of GDP regularizer, compared with adversarial debiasing, for regression and classification tasks in tabular and graph benchmarks 1. We evaluate the effectiveness and expansibility of GDP. |
| Researcher Affiliation | Academia | Zhimeng Jiang1, Xiaotian Han1, Chao Fan1, Fan Yang2, Ali Mostafavi1, and Xia Hu2 1Texas A&M University, 2Rice University |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Codes are available at https://github.com/zhimengj0326/GDP |
| Open Datasets | Yes | UCI Adult dataset 5 contains more than 40, 000 individual information from 1994 US Census. 5https://archive.ics.uci.edu/ml/datasets/adult. The Crime dataset 7 includes 128 attributes for 1, 994 samples from communities in the US. 7https://archive.ics.uci.edu/ml/datasets/communities+and-crime. The temporal graph data, provided by data intelligence company Cuebiq (Cuebiq, 2021)..., citing Cuebiq. Data for good: Location intelligence for good is our contribution to the scientific community, 2021. URL https://www.cuebiq.com/about/data-for-good/. |
| Dataset Splits | Yes | In each trial, the dataset is randomly split into a training, validation, and test set with 50%, 25%, and 25% partition, respectively. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, memory) used for running experiments were mentioned in the paper. |
| Software Dependencies | No | The paper mentions several models and frameworks (e.g., selu networks, GCN, GAT, SGC, TGAT) and their respective citations, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | Model: We adopt two-layer selu networks model (Klambauer et al., 2017) with hidden size 50 and report the mean prediction performance and GDP with 5 running times. Model: We use three graph neural network backbones... We train GNN with 200 epochs with 5 running times... The hyper-parameter λ in Eq. (5) controls the tradeoff between prediction performance and GDP. |