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..

A Single-Loop First-Order Algorithm for Linearly Constrained Bilevel Optimization

Authors: Wei Shen, Jiawei zhang, Minhui Huang, Cong Shen

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments corroborate our theoretical findings and demonstrate the practical efficiency of the proposed SFLCB algorithm. Simulation code is provided at https://github.com/Shen Group/SFLCB. 1 Introduction ... 6 Experiments In this section, we evaluate the performance of our SFLCB algorithm on three tasks: a toy example, hyperparameter optimization for SVM, and a transportation network design problem. These experiments demonstrate the practical effectiveness and efficiency of SFLCB.
Researcher Affiliation Collaboration Wei Shen University of Virginia EMAIL Jiawei Zhang University of Wisconsin-Madison EMAIL Minhui Huang Meta EMAIL Cong Shen University of Virginia EMAIL
Pseudocode Yes Algorithm 1 SFLCB Input: δ, ρ1, ρ2, ηx, ηy, ηz, ηv, ηu, T Initialize: x0 X, y 0, z 0 Py, u0, v0 Rdh for t = 0, 1, ..., T 1 do ut+1 = ut + ηu(h(xt, yt) αt) vt+1 = vt + ηv(h(xt, zt) βt) xt+1 = xt ηx x K(xt, y t, z t, ut+1, vt+1) y t+1 = ΠPy{y t ηy y K(xt, y t, z t, ut+1, vt+1)} z t+1 = ΠPy{z t + ηz z K(xt, y t, z t, ut+1, vt+1)} end for
Open Source Code Yes Simulation code is provided at https://github.com/Shen Group/SFLCB.
Open Datasets Yes Here we consider the same problem formulation as in [16], which formulates this problem as a coupled linearly constrained BLO problem. We conduct experiments comparing our SFLCB algorithm with GAM [46], LV-HBA [50], BLOCC [16], and Bi C-GAFFA [49] on the diabetes dataset [6]. [6] Dheeru Dua, Casey Graff, et al. UCI machine learning repository. 2017.
Dataset Splits Yes The experiments are conducted across 10 different random train-validation-test splits, and the average results along with one standard deviation are reported in Figure 2.
Hardware Specification Yes The experiments on the toy example and hyperparameter optimization for SVM are conducted on an AMD EPYC 9554 64-Core Processor. The experiments on transportation network design are conducted on an Intel(R) Xeon(R) Platinum 8375C CPU.
Software Dependencies No We adapt and modify the code from [16].
Experiment Setup Yes For the toy example, we set the hyperparameters for our algorithm as δ = 0.1, ηx = ηy = ηz = ηu = ηv = 0.01, ρ1 = ρ2 = 1. ... For our algorithm SFLCB, we set δ = 0.01, ηx = ηy = ηz = ηu = ηv = 0.001, ρ1 = ρ2 = 0.01. ... For the 3 nodes setting, we set the hyperparameters of our method SFLCB as: δ = 0.1, ρ1 = ρ2 = 1000, ηx = ηy = ηz = ηu = ηv = 3e 4, and T = 30000. For the 9 nodes setting, we set the hyperparameters of our method SFLCB as: δ = 0.25, ρ1 = ρ2 = 50, ηx = ηy = ηz = ηu = ηv = 3e 5, and T = 300000.