The use of Keras to model flow control mechanisms with recurrent neural networks

C. JARNE; R. LAJE

La Habana

Otro; School and Workshop on Statistical Physics Approaches to Systems Biology.; 2019

Facultad de Física de la Universidad de La Habana -RED INFERNET

There are some recent advances in algorithms, software, and technology which are related to dynamical systems and machine learning that has not yet been applied in the field of Computational Neurosciences, especially regarding computing. Main goal of present work is to present a simple model and a framework to perform a set of tasks related to flow control and discuss the obtained results using this new framework. We proposed a network topology, constraints on the parameters and a training method in order to archive those tasks and we discussed the scope of the results obtained. Trained networks serve as a source of mechanistic hypotheses and as a testing ground for data analyses that link neural computation to behavior. RNN are a convenient proxy for biological circuits and a valuable platform for theoretical investigation. We focus on the study of the following list of tasks regarding the processing of stimulus as temporal input.