Section: New Results

Overview

Though our view is systemic, our daily research activities are also concerned with the design, at a given scale of description, of models of neuronal structures, each concerned with a specific learning paradigm. Of course, a major challenge is to integrate these elements in a systemic view, i.e. to put a specific emphasis on the way each neuronal structure communicates with the rest of the system and to highlight how its learning paradigms interact with other memory systems.

Among the numerous loops involving the brain, the body and the environment, a basic grid of description corresponds to distinguish “perception aspects of loops”, the goal of which is to extract from the inner and outer world sensory invariants helpful to identify and evaluate the current state and to make predictions from previous learning, and “action aspects of loops”, the goal of which is to rely on this sensory and emotional information to decide, plan and trigger actions for the benefit of the body.

This year, our team was engaged on the following topics: Concerning perception aspects of loops, we published original models of the amygdala and of the hippocampus and considered their role in pavlovian conditioning and their evaluation as classical models in machine learning. Concerning action aspects of loops, in addition to a critical analysis of the current views of the interactions between the prefrontal cortex and the basal ganglia [15] , we have proposed an original model for the formation of habits and have also studied related theoretical problems in machine learning, for data representation. Finally, we also report here more methodological achievements, corresponding to the design of algorithmic ersatz of cerebral subsystems.