Please login first
SMANN: AutoML Screening Model of Artificial Neural Networks for Brain Connectome
1 , 2 , * 1 , 1 , * 3
1  Department of Computer Sciences, University of A Coruña (UDC), A Coruña, 15071, A Coruña, Spain.
2  Department of Molecular and Cellular Pharmacology, Miller School of Medicine and Center for Computational Science, University of Miami, FL 33136, Miami, USA.
3  Department of Organic Chemistry II, University of the Basque Country UPV/EHU, 48940, Bilbao, Spain.


We can represent the brain as a Brain Connectome Network (BCN) formed by ni brain cortex regions interacting with others (Lij = 1) or not (Lij = 0). The large number of links to be study and their complex connectivity patterns made difficult to select the appropriate topology in order to predict them with Artificial Neural Networks (ANNs) algorithms. In this context, Automated Machine Learning (AutoML) techniques may help non-experts to select, train, validate, and use automatically the correct algorithms. In this work, we developed a new ÁutoML Screening Model for ANN (SMANN) algorithm to solve this problem. We can quantify topological (connectivity) information of both the complex networks under study and a set of ANNs trained using Shannon measures. The SMANN model presented >85% of accuracy for 52690 outputs of 10 different ANNs for 52690 BCN links.

Keywords: Artificial Neural Netoworks; Brain Cortex; Connectome; Linear Discriminant Analysis