Network Inference of Dynamic Models by the Combination of Spanning Arborescences

International audienceIn this paper, we tackle the problem of generative learning of dynamic models from”fat” time series data (high #variables/#individuals ratio), leading to a high sensitivity of learnedmodels to the dataset noise. To overcome this problem, we propose a method computing a mixtureof many highly biased but optimal spanning arborescences obtained from many perturbed versionsof the original dataset, introducing variance to counterbalance the strong arborescence bias. Themethod is theoretically at the boundary between structure oriented Bayesian model averagingand recent work on density estimation using mixtures of poly-trees through a perturb and combineframework, transposed to a dynamic setting. In practice, preliminary results on the recent DREAMD8C1 challenge are promising