Iterative Connecting Probability Estimation for Networks

Authors: Yichen Qin, Linhan Yu, Yang Li

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We establish desirable theoretical properties for our method, and further justify its superior performance by comparing with existing methods in simulation and real data analysis. 5 Experiments To evaluate the effectiveness of our proposed method, we compare its performance with several popular estimation methods using simulated networks with different features...
Researcher Affiliation Academia Yichen Qin University of Cincinnati qinyn@ucmail.uc.edu Linhan Yu Renmin University of China yulinhan47@foxmail.com Yang Li Renmin University of China yang.li@ruc.edu.cn
Pseudocode Yes Algorithm 1 Iterative connecting probability estimation method and Algorithm 2 Tuning parameters selection of ICE via edge cross-validation
Open Source Code Yes We include the code and data in the supplemental material and will publish them online later.
Open Datasets Yes We analyze a human brain projectome dataset from an experiment of Beijing Normal University in China (Yan et al., 2009)2. The dataset is available on https://Neuro Data.io/, a platform that enables large-scale neurodata storing, analyzing, and modeling.
Dataset Splits Yes The tuning parameters can be selected by network cross-validation. Randomly sample a subset of edges from E with probability p to obtain the training set of the edges Etrain. Let Eval = E Etrain denote the validation set.
Hardware Specification No [No] Since our method is computationally feasible for networks with moderate size, we omit this part for brevity.
Software Dependencies No The paper does not provide specific software dependencies (e.g., library names with version numbers) needed to replicate the experiment.
Experiment Setup Yes As to the size of the similar vertex set s, Zhang et al. (2017) set s = C(n log n)1/2 for each vertex i, where C is recommended set as 1. The performance of the combination (Cit = 0.2, Cest = 1) is the most competitive and even comparable to that of Oracle. Input: observed adjacency matrix A; initial connecting probability estimate b P(0); neighborhood size s; threshold δ0 > 0.