Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Individually Fair Diversity Maximization
Authors: Ruien Li, Yanhao Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments on real-world and synthetic datasets demonstrate that the proposed algorithms generate solutions that are individually fairer than those produced by unconstrained algorithms and incur only modest losses in diversity. |
| Researcher Affiliation | Academia | Ruien Li School of Data Science and Engineering East China Normal University Shanghai, China EMAIL Yanhao Wang School of Data Science and Engineering East China Normal University Shanghai, China EMAIL |
| Pseudocode | Yes | Algorithm 1: IFRGENERATE Input: Fairness parameter α Output: A set of individual fairness regions for given parameters α, k Initialize the set of covered points Z and the set of centers of selected balls C while Z = P do... |
| Open Source Code | Yes | Our code and data are publicly available at https://github.com/HonokaKousaka/IFDM. |
| Open Datasets | Yes | In the experiments, we used three public real-world data sets and one synthetic data set. The basic information for each data set is shown in Table 1. For Movie Lens, the user vectors are obtained through matrix factorization using LIBMF [11]. We randomly sampled 1,000 points from each data set for evaluation. Table 1: Statistics of data sets in the experiments, where n is the number of data points and dim is the dimensionality. Dataset Description n dim Celeb A [26] Features for celebrity images extracted by VGG16 202,599 25,088 Glo Ve [37] Global vectors for word representation 400,000 100 Movie Lens [18] User vectors obtained from the rating matrix 162,541 50 Gaussian Gaussian blobs by make_blobs in scikit-learn [36] 1,000,000 20 |
| Dataset Splits | No | We randomly sampled 1,000 points from each data set for evaluation. |
| Hardware Specification | Yes | All experiments were carried out on a server with an Intel(R) Xeon(R) Gold 6134 CPU @3.20GHz (2 processors) and 128GB RAM running Windows Server 2019 Datacenter. |
| Software Dependencies | No | The algorithms were implemented in Python 3. Our code and data are publicly available at https: //github.com/Honoka Kousaka/IFDM. ... We implemented all algorithms in Python 3 and used the Gurobi optimizer to solve ILPs in FMMD-S. |
| Experiment Setup | Yes | For max-min diversification under individual fairness constraints, we implemented FMMD-S [46]... and fixed ε = 0.05 for FMMD-S. For max-sum diversification under individual fairness constraints, we implemented the local search algorithm in [2]... and set ε = 0.05 in the algorithm. ... We fixed α = 1 in all experiments so that each algorithm makes the best effort to ensure individual fairness. |