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..
Vertical Federated Feature Screening
Authors: Huajun Yin, Liyuan Wang, Yingqiu Zhu, Liping Zhu, Danyang Huang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Simulation studies In this section, we adopt numerical results on simulated datasets to demonstrate the effectiveness of VFS. 6 Real data analysis To illustrate the performance of VFS in real data applications, we first consider four publicly available real-world datasets: Activity [1], Gina [32], p53 Mutants [42], and RNA-Seq [25]. |
| Researcher Affiliation | Academia | 1Center for Applied Statistics and School of Statistics, Renmin University of China, Beijing, China 2Bigdata and Responsible Artificial Intelligence for National Governance, Renmin University of China, Beijing, China 3School of Statistics, University of International Business and Economics, Beijing, China 4Institute of Statistics and Big Data, Renmin University of China, Beijing, China Correspondence to: Liyuan Wang <EMAIL> and Danyang Huang <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Vertical Federated Feature Screening (VFS) Input: Feature matrix X Rn p, label series Y Rn, subsample ratio s, initial group size s0, decay coefficient ρ, # features of stage switching p , screening threshold δ. Output: Retained feature matrix X Rn d. for i = 1 to n0 do |
| Open Source Code | Yes | To facilitate the reproducibility of our findings, we will provide the essential code for the VFS algorithm framework as supplemental material. |
| Open Datasets | Yes | To illustrate the performance of VFS in real data applications, we first consider four publicly available real-world datasets: Activity [1], Gina [32], p53 Mutants [42], and RNA-Seq [25]. |
| Dataset Splits | Yes | 80% of the data is randomly split for training and 20% for testing. Accuracy and AUC are both evaluated on the test set. |
| Hardware Specification | Yes | We repeated each experiment M = 200 times on a machine equipped with an Intel(R) Xeon(R) Gold 5320 CPU @ 2.20GHz and 72 GB of RAM, and evaluated the screening performance from two perspectives: (1) statistical effectiveness, measured by the Positive Selection Rate (PSR = | ˆF F|/|F|) and the False Discovery Rate (FDR = | ˆF F|/| ˆF|), and (2) computational efficiency, evaluated by the average total computation time denoted as Time. |
| Software Dependencies | No | The paper mentions homomorphic encryption schemes (Paillier [58] and CKKS [13]) but does not provide specific version numbers for any software libraries, programming languages, or other tools used for implementation. |
| Experiment Setup | Yes | The default simulation setting is as follows: sample size n = 10000, number of features p = 2000, number of true features m = 20, initial group size s0 = 100, decay coefficient ρ = 0.5 and number of selected features after screening d = 100. ... Specifically, we investigate the impact of hyperparameter on the performance of VFS, including subsampling ratios s = 1, 2, 3, 4, 5, initial group size s0 = 10, 20, 50, 100, 200, and number of retained features after screening d = 50, 100, 150, 200, 250. |