Connectivity, dynamics, and biological constraints of recurrent neural networks used to model the brain

C. JARNE

Austin Texas (Participación virtual)

Conferencia; 33rd IUPAP Conference on Computational Physics; 2022

The University of Texas at Austin

Abstract: Some brain regions, especially the cortex, show great recurrence in their connections, even in early sensory areas. Several approaches and methods based on trained networks have been proposed to model and describe these regions. To archive that, it is essential to understand the dynamics behind the models, in particular, because they are used to explain experimental results and build different hypotheses about the functioning of the brain. Here, we discuss connectivity patterns, dynamics, and biological constraints of recurrent neural networks considered when training such models in different decision-making and temporal tasks. In addition to theoretical elements, it is crucial to have a software framework that allows researchers to easily test different hypotheses and constraints. Here, a simple framework for training networks based on Tensorflow and Keras is presented.