Finding Wasserstein Ball Center: Efficient Algorithm and The Applications in Fairness
Authors: Yuntao Wang, Yuxuan Li, Qingyuan Yang, Hu Ding
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we conduct a set of experiments on both synthetic and real-world datasets, demonstrating the computational efficiency of our algorithm, and showing its ability to provide more fairness for input distributions. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, University of Science and Technology of China, Hefei, China 2School of Artificial Intelligence and Data Science, University of Science and Technology of China, Hefei, China. Correspondence to: Hu Ding <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Solver for ( AR A ) y = f ... Algorithm 2 Predictor-Corrector Inner Point Method for Linear Programming (WBC-LP) |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its source code, nor does it include a link to a code repository for the methodology described. |
| Open Datasets | Yes | (3) The third experiment further illustrates the performance on a real-world dataset Fair Face (Karkkainen & Joo, 2021) with diverse racial representation. ... (4) The fourth experiment on 3D point-cloud gives a visual example to the fairness of WBC as a barycenter. We take the Shape Net Core-55 dataset (Chang et al., 2015), with each cloud containing 2048 points in R3. |
| Dataset Splits | No | The paper describes how input data for the experiments was generated or selected, such as partitioning experimental distributions into families or choosing 700 images for different racial groups. However, it does not provide specific training/test/validation dataset splits in the context of machine learning model training. |
| Hardware Specification | Yes | All the experiments are implemented on a workstation, Intel(R) Core(TM) i59400 CPU @ 2.90GHz and 8GB for RAM, equipped with win64 Windows 11+.0. |
| Software Dependencies | Yes | The baseline we choose is Gurobi Optimizer version 11.0.0 (academic license). |
| Experiment Setup | Yes | For these two experiments, we generate random datasets in Euclidean space, and the weights of (q(t) 1 , ..., q(t) m ) in each distribution P(t) are generated uniformly at random. ... In L1 space of dimension 100, we sample the 30 supports uniformly in a cube with side length 3, then translate 3 supports by adding 1100, and translate 5 supports by adding 2 1100, we have var(WD)=58.35 for WB , while var(WD)=418.61 for WBC. |