Braincraft associated team

What are the processes by which animals and humans select their actions based on their motivations and on the consequences of past actions? This is a fundamental question in neuroscience, with implications to ethology, psychology, economics, sociology and computer science. The basal ganglia are a group of interconnected subcortical nuclei which have been the subject of intensive studies in the past years, both at the computational, psychological, anatomical and biological levels (Alexander, 1986; Mink, 1996; Gurney et al, 2001, Leblois 2006). From study of the basal ganglia’s dysfunctions (e.g. neurodegenerative diseases), these nuclei are believed to play a prominent role in action selection, choosing, among possible actions, to trigger the one with the best expectancy of achievement, with regard to the goal to be fulfilled. The place for the representation of actions and of their contingencies with regard to motor and sensory constraints is the prefrontal cortex, in charge of the temporal organization of behavior (O’Reilly et al, 2006). The prefrontal cortex is consequently at the crossroad of two kinds of loops, with other cortical regions giving sensorimotor constraints from the environment and with the basal ganglia giving reinforcement constraints in terms of goal achievement.

A central question in cognitive computational neuroscience is to understand how cognitive control emerge from these regions and elaborate criteria, in the process of making an overall decision, which often require weighting various tradeoffs (Kolling, 2012; Fiorillo C.D., 2003). A particularly important question in this perspective is to know more precisely the nature and the moment of the contribution of the basal ganglia in this process. As often in this domain, this question evokes two main aspects: (i) information representation: which kind of information is represented in these structures and how it is exchanged between them? (ii) learning: what kind of learning is performed in each of these structures and what is its impact on the others?

Investigators

First year (2015)

This first year of the associated team has given us the opportunity to exchange with our US partners on the various topics mentioned in the program. This convinced us to start to explore the role of the OFC in the decision making process and the dynamic valuation of stimuli. Furthermore, we (on the French side) started to investigate the Emergent software platform in order to check if it can be used for our own models that differ a bit in their functional hypothesis and basic computational unit. In the same time, we’re also investigating how to back port some the Emergent models into the Python/Numpy framework such as making the model easily shareable and replicable.

Record of activities

Second year (2016)

We mentioned last year that we do not have yet a common platform for sharing models. The US team is mostly using the Emergent framework (that they are developing) while the French team is mostly using the Python language associated with the classical scientific stack (numpy, scipy, matplotib, cython). This incompatibility prevents us from sharing models or components and represents the main barrier for deeper collaborations. Fortunately, Fabien Benureau joined the French partner as a post-doc and has a great experience in Python. In cooperation with Jessica Mollick, he started the replication of two models (PVLV and bvPVLV) that are representative of the kind of models the US partner is doing. This work is not yet finished but early results tend to show this will be finished by next year and this should help us to exchange both model parts and computational principles underlying those models.

In the meantime, we are also exchanging publications, materials and experience on the modeling of the medial prefrontal cortex, a structure known to be involved in the estimation of the emotional value of surrounding objects and in the evaluation of motivation to act (need, fatigue). Thomas Boraud is also active in the discussion since he is currently experimenting electrophysiology in this region in monkeys. One important goal of this discussion is to precisely define a well-defined and motivated experimental protocol to be presented for a proposition of common ANR-NSF project. Both teams are also working on the precise understanding of reward prediction errors in a neuronal structure called VTA (Ventral Tegmental Area), known to be of utmost importance for respondent as well as instrumental conditioning. Furthermore, we’re also investigating respective modeling works to compare similar mechanisms and to understand when they are complementary. These exchanges were particularly fruitful with respect to models of dopamine release even if we disagreed on some specific mechanisms.

Record of activities

References