CORTEX: Neuromimetic intelligence

Can we model and reproduce neuron computing, that is to say the functioning of the brain ?

Experimental olfactory perception platform

How does the brain coordinate all the information it receives in order to make decisions and activate the motor functions of the body ? Understanding this functioning is a major challenge in neuroscience. Project CORTEX has decided to meet this challenge by attempting to model simple functions in order to reproduce neuron computing on several levels: the functioning of a single neuron, computing on neuron networks, a sensory function, actions and so on. If a global model of the biological brain remains utterly inaccessible, approaches using elementary circuits are nonetheless getting closer.

The team is working with two robots in the framework of a European project called MirrorBot started in June 2002. The robots are equipped with wheels and one arm. They are used to better understand movements, going around obstacles and sensorymotor coordination-the fact of grasping an object that was recognized for example. Such simple tasks that humans carry out "without thinking about it" are a lot more complex than it seems, and implementing neuron networks in the robots made it possible to identify qualities of the human brain: robustness and extreme adaptability. Even when it is partly damaged or when the object to be grabbed or the necessary movements are not exactly the same, the human brain automatically compensates for it and carries out the planned task. Today's models are far from conferring the same efficiency to robots. This is one of the major stakes of Cortex research.

CORTEX applied its work to the modeling of one of the "simplest" sensory functions, olfaction. The NOSE research initiative involving computer scientists from LORIA-INRIA, physicists from the Henri Poincaré University in Nancy and biologists from the Cognitive Sciences Institute in Lyons resulted in the construction of an "electronic nose".

Olfaction however only involves a few hundred neurons, whereas most cerebral functions entail the participation of millions of neurons. An olfaction model can be made in real size, but other functions, such as the sensorymotor functions, require a simplified approach using groups of neurons in order to remain workable. This is why CORTEX is also working on circuits composed of one hundred to one thousand neurons, in addition to its work on the modeling of the functioning of one neuron. Additionally, computations are done in parallel in the human brain. We do not wait for a neuron to be finished with a computation to start with the next neuron. Cortex develops parallel computing to approximate the delicacy of biological computing.

The practical applications of this work are wide ranging. Understanding brain behavior makes it possible to advance in the treatment of certain brain malfunctions. It is, for example, possible to model neuron network lesions. Going beyond neuroscience, neuron computing turns out to be a model for a new kind of mathematical computing that is very efficient to process or use vast quantities of data. It is used to control machines or complex systems such as telecommunication systems, among other tasks.

Practical information

Project launched in 2000
Partners: INRIA, CNRS, the Henri Poincaré and Nancy 2 Universities and INPL.

Research directions

• Behavioral neuromimetic modeling for robot autonomous navigation: development and interfacing of associative cortex models (sensorymotor coordination), frontal cortex models (temporal organization of behavior) and hippocampus (localization).
• Elementary neuromimetic modeling for the binary functioning of the neuron (action potential emission or spikes).
• Adaptation of continuous classical neuron models to database exploration problems, with a view toward neurosymbolic integration (knowledge extraction and integration). This work leads to the exploration of other computing directions such as parallel architectures, autonomous robots and more generally, specialized circuits for embedded systems.

The team
Ten research scientists (INRIA, CNRS) and university professors (Universities of Nancy and Strasbourg).
A dozen doctoral candidates and post-docs.

Scientific Head
Frédéric Alexandre

Back to top