Efficient Network Automatic Relevance Determination

Authors: Hongwei Zhang, Ziqi Ye, Xinyuan Wang, Xin Guo, Zenglin Xu, Yuan Cheng, Zixin Hu, Yuan Qi

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the performance of the proposed methods on synthetic data using True Positive Rate (TPR), defined as TPR = TP TP+FN, and False Positive Rate (FPR), defined as FPR = FP TN+FP, where TP, FP, TN and FN represent true positives, false positives, true negatives, and false negatives, respectively. We run each experiment 10 times with different random seeds and report the average. For Aging phenotype data, where true labels are unavailable, we use the Jaccard index to measure the overlap of biological associations identified by different algorithms. We also highlight the efficiency of the NARD and its variants. On the TCGA data, we focus on the effectiveness of NARD in identifying biological associations.
Researcher Affiliation Academia 1Artificial Intelligence Innovation and Incubation Institute, Fudan University, Shanghai, China 2School of Data Science, Fudan University, Shanghai, China 3Shanghai Academy of Artificial Intelligence for Science, Shanghai, China 4Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, United Kingdom 5Eberly College of Science, Pennsylvania State University, PA, United State 6Zhongshan Hospital, Fudan University, Shanghai, China. Correspondence to: Zixin Hu <EMAIL>, Yuan Qi <EMAIL>.
Pseudocode Yes Algorithm 1 NARD Input: Input data X, Y , ϵ Output: Estimated α, V , V 1, W 1: Initialize α elements, V ˆVMLE. 2: for t 1 to T do 3: Compute Σ (K + XX ) 1. 4: Compute µ Y X Σ. 5: Update V according to (15). 6: V, V 1 glasso(V, λ). 7: Update α according to (16), (18) depending on whether the Gamma prior is included. 8: if max | ( 1 α)| ϵ then 9: Break. 10: end if 11: Update W µ. 12: end for
Open Source Code No The paper provides links and descriptions for open-source code of *baseline* methods used for comparison (MRCE, CAPME, HS-GHS, JRNS), but does not provide concrete access to the source code for the novel methods (NARD and its variants) proposed in this paper.
Open Datasets Yes On the TCGA data, we focus on the effectiveness of NARD in identifying biological associations. In addition, we conducted experiments on financial and air quality datasets to further demonstrate the versatility of our method; the detailed experimental results are provided in Appendix F.2. All experiments were performed on 32 Intel(R) Xeon(R) Platinum CPUs. TCGA cancer data (Weinstein et al., 2013) Kaggle s air quality dataset5 https://archive.ics.uci.edu/dataset/501/beijing+multi+site+air+quality+data
Dataset Splits Yes To select the optimal λ, we employ a 5-fold cross-validation procedure. The dataset is partitioned into 5 disjoint subsets, and in each iteration, 1 subset is held out as the validation set while the remaining 4 subsets are used for model estimation. The objective function for selecting λglasso is defined as: We run each experiment 10 times with different random seeds and report the average.
Hardware Specification Yes All experiments were performed on 32 Intel(R) Xeon(R) Platinum CPUs.
Software Dependencies No The paper does not explicitly provide specific version numbers for any software dependencies or libraries used in their implementation.
Experiment Setup Yes To select the optimal λ, we employ a 5-fold cross-validation procedure. The dataset is partitioned into 5 disjoint subsets, and in each iteration, 1 subset is held out as the validation set while the remaining 4 subsets are used for model estimation. The objective function for selecting λglasso is defined as: Algorithm 1 NARD Input: Input data X, Y , ϵ Output: Estimated α, V , V 1, W 1: Initialize α elements, V ˆVMLE.