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..
Fairness-aware Bayes Optimal Functional Classification
Authors: Xiaoyu Hu, Gengyu Xue, Zhenhua Lin, Yi Yu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our theoretical findings are complemented by extensive numerical experiments on synthetic and real datasets, highlighting the practicality of our designed algorithm. |
| Researcher Affiliation | Academia | School of Mathematics and Statistics, Xi an Jiaotong University Department of Statistics, University of Warwick Department of Statistics and Data Science, National University of Singapore Department of Statistics, University of Warwick |
| Pseudocode | Yes | Algorithm 1 Fair Functional Linear Discriminant Analysis classifier. |
| Open Source Code | Yes | We have submitted code including those generating all the numerical results in this paper. |
| Open Datasets | Yes | For the real dataset, we use the 2005-2006 National Health and Nutrition Examination Survey data (CDC, 2006), where the sensitive attribute is race and the classification task is to determine if an individual is under 20 or over 50 years old based on the quantile function of intensity values. ... The real dataset is obtained from https://wwwn.cdc.gov/nchs/nhanes/Continuous Nhanes/Default.aspx?Begin Year=2005. |
| Dataset Splits | Yes | The final dataset consists of 3252 instances, which we randomly split into equal-sized training and test subsets. ... Truncation levels are selected via 5-fold cross-validation, specifically by minimising the average classification error associated with the unconstrained classifier. ... One subset is used to estimate bηa and bπa,y, while the other is used to estimate the threshold bτ. ... To mitigate the randomness caused by random splitting, we adopt a cross-fitting approach and define the final probabilistic classifier as the average bf = ( bf1 + bf2)/2. |
| Hardware Specification | Yes | Experiments were conducted on a server equipped with an Intel(R) Xeon(R) Platinum 8280 CPU @ 2.70GHz (28 cores) and 503GB of RAM. |
| Software Dependencies | Yes | We implemented all methods in R (version 4.3.1). |
| Experiment Setup | Yes | Truncation levels are selected via 5-fold cross-validation, specifically by minimising the average classification error associated with the unconstrained classifier. Fair-FLDA: calibration constant set to 0; Fair-FLDAc: calibration constant set to min{ p 2 log(1/ρ)/n, δ}, with ρ = 0.05. For the simulation results, we generate (Y, A) {0, 1} 2 according to the distributions P(A = 1) = 0.7, P(Y = 1|A = 0) = 0.4 and P(Y = 1|A = 1) = 0.7. Given Y = y and A = a, generate the functional covariate Xa,y(t) as Xa,y(t) = µa,y(t) + P50 k=1 ζa,kφk(t), where φk(t) = 2 cos(kπt), ζa,k N(0, λa,k), λ0,k = k 2, λ1,k = 2k 2, and the mean functions are specified as follows, µ0,0 = µ1,0 = 0, µ0,1(t) = k=1 0.8( 1)kk βφk(t), µ1,1(t) = 2( 1)kk βφk(t). Let β = 1.5 and n = 1000. |