Latent Process Models for Functional Network Data

Authors: Peter W. MacDonald, Elizaveta Levina, Ji Zhu

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

Reproducibility Variable Result LLM Response
Research Type Experimental We derive a gradient descent estimation algorithm, establish theoretical guarantees for recovery of the low dimensional structure, compare our method to competitors, and apply it to a data set of international political interactions over time, showing our proposed method to adapt well to data, outperform competitors, and provide interpretable and meaningful results. Keywords: Latent space model, Multilayer, Multiplex, Dynamic network, B-spline
Researcher Affiliation Academia Peter W. Mac Donald EMAIL Department of Statistics & Actuarial Science University of Waterloo Waterloo, ON N2L 3G1, Canada Elizaveta Levina EMAIL Ji Zhu EMAIL Department of Statistics University of Michigan Ann Arbor, MI 48109-1107, USA
Pseudocode Yes Algorithm 1: Concurrent gradient descent algorithm. Algorithm 2: Local average embedding initialization algorithm. Algorithm 3: Sequential gradient descent algorithm.
Open Source Code Yes FASE is implemented in an R package fase available on CRAN.
Open Datasets Yes As an application to real functional network data, we apply FASE to data collected by the Integrated Crisis Early Warning System (ICEWS) (Lautenschlager et al., 2015). ... URL https://doi.org/10.7910/DVN/28117.
Dataset Splits Yes We then uniformly select a random snapshot index x k [0.25, 0.5] (to avoid boundary effects) and remove it along with M snapshots immediately before and after x k, for M = 0, 1, ..., 10. That is, we treat the 2M +1 network snapshots closest to the selected snapshot in the index space as missing.
Hardware Specification Yes As an example benchmark, fitting FASE with n = m = 100, d = 2 and q = 10 to data generated as in scenario (i) takes about 4 seconds on a computer with an Apple M2 Pro Chip and 16GB RAM.
Software Dependencies No FASE is implemented in an R package fase available on CRAN. ... Implementations of both concurrent and sequential FASE estimators are available in the R package fase.
Experiment Setup Yes In these simulations, we set the convergence criterion for Algorithm 1 to the relative decrease in the objective function dropping below 10 5. ... In the coordinate descent scheme, we initialize d = 1, and perform alternating minimization for q and d by evaluating NGCV on a grid and treating the other as fixed.