Resumen
The main challenge of this project is to surpass current limitations of learning algorithms in AI and
their application to dynamical systems in Systems Neuroscience. To address these questions, the
project is divided in two main Objectives:
In Objective 1, the NeuroSystems-AI learning algorithm will be developed as the product of several interacting neurobiological mechanisms. We will evaluate the goodness of the resulting algorithm
through cycles of optimization + benchmarking over standard datasets in the Machine Learning field.
In Objective 2, multi-areal RNNs of spiking neurons will be implemented according to the guidelines of the NeuroSystems-AI learning algorithm. We will apply the resulting RNN models to study
neural circuit dynamics underlying goal-directed behaviors, in particular to better understand how the
brain learns to prioritize the relevant information dynamically, and to unravel the learning process by
which generic sensory-response associations are conformed in fronto-striatal sub-circuits.
