Generative Learning of Dynamic Structures using Spanning Arborescence Sets
Anthony Coutant, Céline Rouveirol · HAL (Le Centre pour la Communication Scientifique Directe) · 2019
Motivation: We focus on the problem of learning generative Gene Regulatory Network structures from scarce gene expression time series, where the (#variables/#individuals) ratio is high. Results: We propose the ELSA method computing a composite model using Bayesian Model Averaging from optimal spanning arborescences built from perturbed versions of the original dataset. We introduce various strategies to build composite from component models, including the use of both high and low ranked model traits to discriminate models, and validate them on the recent DREAM D8C1 challenge.