Connectionists: PhD position in computational neurosciences at Nancy, France:

[Yann Boniface]

PhD position in computational neurosciences at Nancy, France:

In the framework of the MAPS project (Mappings, Adaptation, Plasticity 
and Spatial computation) funded by the French National Research Agency, 
we are opening a PhD position on the motivated learning of 
spatio-temporal inner structuration of spiking neural networks.

This project gathers four French partners:

- INRIA-CNRS Nancy university
   http://cortex.loria.fr & http://maia.loria.fr/

- Laboratoire Mouvement et Perception,  Marseille
- Institut de Neurosciences Cognitives, Marseille
- Laboratoire d'InfoRmatique en Image et Systèmes d'information, Lyon

The overall goal of the project is to carry out an etensive study of the 
superior colliculus using various experiments and recording technics in 
order to understand the underlying principles that should lead to the 
design of a realistic neural model.


The candidate will have a Master in Computer Science or Applied 
Mathematics, preferably with insights or interest in Neurosciences 
thematics.

Applications including a CV, should be sent by email to Y.Boniface and 
A.Dutech (Yann.Boniface at loria.fr,Alain.Dutech at loria.fr)


------------------
subject :

Study of the functional properties of neural networks temporal coding

Based on neural population mean activity, the Cortex team has designed 
models of selective visual attention based on distributed, adaptive and 
numerical computations. However, while those models exhibit strong 
attentional properties, they fail at properly integrating time, mainly 
because of their discrete temporal nature. To cope with this problem, a 
model of visual attention needs to be designed using the spiking neurons 
paradigm and we would like to push further in this direction by 
investigating temporal coding and functional properties resulting from 
learning with regards to the overall network stability. In the long run, 
we would like to investigate anticipation mechanisms and understand what 
is the benefit of such temporal coding.
We want more particularly to explicit the formation of spatially 
structured and temporally operant connectivity patterns within the 
neural substrate with forward and lateral flow. Local Hebbian synaptic 
rules (temporal or not) will be thoroughly studied in conjunction with 
other synaptic mechanisms, like short term depression and synaptic 
traces. Those approaches will be compared with higher level 
extraction/competition processes that allow the formation of specialized 
receptor fields. One of the issue is to fill the gap between the 
classical self-organized feature-extraction models and realistic 
synaptic based adaptation processes. We hope in particular to clarify 
the lateral specialization processes that could allow the formation of 
specialized bubbles of activity.
 From a broader point a view, this work is to be placed in the framework 
of multi-modal integration: human cognition is based on a unified view 
of the world that is build upon perceptions coming from all the 
different modalities (vision, olfaction, proprioception, etc.). Even if 
there are dedicated pathways for each of these modalities, this does not 
result in a fragmented view of the world. In fact, the brain is able to 
integrate all those modalities and to offer a coherent and unified view. 
We would like to be able to build such a unified view of the world and 
this begins by considering first simpler multi-modal associations.