Post-doctoral fellowship

[Nicolas Rougier]

The LORIA laboratory offers a one-year fellowship for a post-doctoral
researcher:

---------------------------------------------------------------------
A connectionist architecture for a multi-level approach of navigation
---------------------------------------------------------------------

We are studying the different kinds of information representations
that may be useful to an autonomous robot seen as a cognitive system.
Recent studies corresponded to integrate data from neurosciences in
connectionist computational models and we studied in this framework how
both procedural, declarative and working memories were identified as
emerging from specific neuronal structures and critical at different
levels of information processing.

For navigational tasks, there exists today an extensive literature on
computational models that are grounded on a core mechanism: the
chaining together of neighboring locations in space. Having a precise
goal to reach, this kind of chaining can then be used through the
equivalent of a backward search to reach the goal from a given place.
Nonetheless, if this strategy can account for quite short
spatio-temporal sequences, it is our belief it cannot account
for a more global and complex navigational behaviour.

Based on our understanding on the role of the different information
representation, we think that going from one place to another distant
one implies scheduling of action at several different levels, both in
time and space. Specifically, we anticipate that a model grounded on
data from prefrontal cortex studies could encode navigational
information at varying levels of spatial and temporal resolutions and
would then be able to perform a precise contention scheduling of the
task. For example, going from one place A1 to another place A2
does not necessarily imply to know that at any given time it is
necessary to go through each of the Ai places. It is more likely
that at any given time only a small subset of the current path is
known while the whole path can be described at a higher resolution (by
using abstract representations like rooms or places for example).

In this framework, we aim at understanding how such multi-level
representations can emerge from a connectionist computationnal model
and how it can promote an efficient navigation based on multi-level
representations. From previous work and collaboration, we have found
that data describing the prefrontal and basal ganglia systems give hints
concerning the design of an efficient system and we would like to go
further this way by designing a computational model for the navigation
of a real robot.
------------------------------------------------------------------------

The candidate must have a PhD in cognitive science, computer science,
or a related discipline and expertise in neural network modelling.
Experience in robotic would be an asset but is not essential.

Deadline for applications is 04/15/2003.
Contact: rougier at loria.fr


Dr Nicolas Rougier.