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..

Bilevel Network Learning via Hierarchically Structured Sparsity

Authors: Jiayi Fan, Jingyuan Yang, Shuangge Ma, Mengyun Wu

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive validation demonstrates NNBLNet s effectiveness across synthetic and real-world scenarios, achieving superior F1 scores compared to competitive methods and particularly beneficial for complex systems analysis through its interpretable bi-level structure discovery. 5 Experiment To comprehensively validate our methodology, we established a dual evaluation framework encompassing both synthetic benchmarks and real-world networks.
Researcher Affiliation Academia 1School of Statistics and Data Science, Shanghai University of Finance and Economics 2Department of Biostatistics, Yale School of Public Health EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Two-Stage Proximal Gradient Descent for NNBLNet Input: Data {xi}n i=1, learning rate η, regularization parameters ζ1 = λ1, ζ2 = λ2, power γ, number of epochs T, tolerance ϵ Output: Estimated parameters {θ, Γj, { (l) j }D l=1}p j=1 Stage 1: Initial Estimation (non-adaptive) Initialize θ, Γj, and (l) j for all j and l for i = 1 to T do
Open Source Code Yes The implementation code is included in the supplementary material and is publicly accessible on Git Hub at https://github. com/mengyunwu2020/NNBLNet.
Open Datasets Yes The Friendship dataset is available at http://www.sociopatterns. org/datasets/high-school-contact-and-friendship-networks/. The Co-authorship dataset is available at https://dl.acm.org/. BRCA and LUAD expression data were obtained from the R package cdgsr, pathway information from the KEGG database was obtained using msigdbr, and interaction information from STRING was obtained via STRINGdb.
Dataset Splits Yes We repeatedly drew 90% subsets of the data from two synthetic and four real-world datasets and re-estimated the networks 100 times.
Hardware Specification Yes All experiments were conducted on a workstation equipped with an Intel Core i7-800H Processor, an Nvidia Tesla A40 GPU, and 64GB of RAM.
Software Dependencies No The paper mentions R packages for data acquisition (cdgsr, msigdbr, STRINGdb) but does not provide specific version numbers for these or for the main development environment (e.g., Python, PyTorch, TensorFlow, etc.) used for the methodology.
Experiment Setup Yes Following the convergence conditions in Theorems 3.6 and 3.7, we set λk = ζk = c n 1/8 for k = 1, 2, with c as a tunable constant. ... Empirically, we recommend c = 0.35, as it achieves satisfactory accuracy across diverse data settings. ... Consistent with common practice, we set γ = 1. Regarding neural network hyperparameters, exploratory experiments (Appendix A.4) showed that a configuration of 1000 training epochs, three hidden layers, and 50 nodes per layer offers an optimal balance between accuracy and computational efficiency.