Project of the Aquitaine Regional Council: Decision making, from motor primitives to action

The aim of this project (partly funding the PhD of Meropi Topalidou) is to investigate decision making at intermediate level in order to establish the link between motor primitives and higher level actions. The question is to understand how continuous complex motor sequences can be dynamically represented as actions such that they can be manipulated to resolve conflict when several actions are possible. This PhD work will require an extensive review of the literature and more specifically literature that promote a global view on decision making. The DANA modeling framework will be used for the design of distributed, numerical and adaptive models using rate based neuron models. The model will ideally be embodied into a simulator or a robotic platform in order to solve a simple taks such as for example, foraging or grasping, with a continuous component at the motor level.

Project of the Department Sciences and Technologies of the University of Bordeaux: Pinokio

In collaboration with school of engineers ENSEIRB and the support of the Department Sciences and Technologies of the University of Bordeaux, we've built a prototype of a motorized lamp equipped with a camera and leds. It can move autonomously and track faces with dedicated algorithms. The goal of this project is to have a dedicated robotic platform to study motor interaction and to investigate decision making in order to establish the link between motor primitives and higher level actions.

Project PEPS of the Idex: Dopamine control of a novel basal ganglia cell-type

The neurotransmitter dopamine (DA) plays a key role in basal ganglia (BG) circuits. However, despite the fundamental importance of DA in those circuits, the electrophysiological effects of dopamine on target neurons are largely unknown. Furthermore, contrary to classical models that only view the globus pallidus (GP) as a relay station of the indirect pathway, our neuroscientist colleagues at IMN have discovered a novel GP cell-type called the Arkypallidal (Arky-GP) neurons that only project to striatum in a very dense way. Arky-GP cells represent a novel BG pathway that might contribute massively to the GABAergic inhibition in striatum. In this project, we would like to explore for the first time whether DA has a direct action on Arky-GP neurons through D2 DA receptors. To do so, this project is based on multidisciplinary approaches that bring together 3 teams of IMN with different but complementary expertise (anatomical, in vivo electrophysiology, optogenetic manipulation, and computational modeling).

Collaboration with the Neurocentre Magendie on parameter optimization: Neurobees

The development of computational models of neurons and networks typically involves tuning of the numerical parameters to fit experimental results. This fitting is necessary to obtain consistent neural activity and therefore consistent action potential genesis and timing which play a key role in neural information encoding. However his task requires the exploration of multidimensional parameter spaces which are rarely accessible to analytical approaches. Moreover, if the parameter tuning can sometimes be manually completed it is more convenient to use automated optimization algorithms at least for two reasons: (i) to apply an homogeneous processing to all the calculation and parameter space exploration which alleviates operator influence and (ii) to avoid a tedious and uncertain result from human operators when the dimensionality increases. In computational neuroscience, the optimization algorithms are often applied to cell scale models to mimic the electrical activity of their biological counterpart. Most of the time, it is necessary for the neuroscientist to quantify biophysical parameters such as dynamic conductances, ionic concentrations or even neuronal structure to understand the neuron dynamic properties. In this field, there is an important need for innovative optimization tools. We have recently developed with neuroscientists of the Bordeaux Magendie Neurocentre, a new multi-agent algorithm in line with ABC (Artificial Bee Colony) paradigm. This algorithm whose principle is based on honeybees food foraging has been successfully applied to several neural modeling optimization problems. We have applied it to several benchmarks and it has shown significantly higher performances in computing optimal parameter values in comparison with the previous optimization tools. A method paper summarizing all these results will be submitted at the begining of 2015.

Collaboration with IMS on GSM signal effects: JNNS (Julia Neural networks Simulator)