learning program
background
 
 
 
 
 
 
Modeling of neuronal circuits
The functioning of neuronal networks depends on complex interactions involving the intrinsic electrical properties of neurons, synaptic dynamics and network architecture (R1). The number of parameters is high and many do not lend themselves to experimental manipulation. A more feasible way to investigate   parameter space is to use computer modeling (R2). An important question for Neurovers-IT is whether computer modeling can mimic the learning capacity of the cultured neuronal networks and, if so, what is the complexity of model needed. Several levels of description are possible, ranging from precise biophysical electrical modeling to analytical representation of network properties.   For precise biophysical modeling the NEURON software package (www.neuron.yale.edu) is often used, while NEST (http://www.nest-initiative.org/) can simulate very large networks of integrate-and-fire units. Both packages have been optimized for parallel computation. Modeling approaches have been successful at simulating the patterns of activity of cultured neurons using modest computer resources. This contrasts with ambitious projects like the Blue Brain Project which plans to simulate the more complex activity in a cortical column, but requires a large supercomputer to do so.
         

Epigenetic & Evolutionary Robotics
Epigenetic robotics   and Evolutionary Robotics have investigated the process through which varied and complex cognitive and perceptual structures emerge from long-term interactions between an embodied system and its physical and social environment. A defining feature of this work is the focus on the pre-cognitive and physical properties that subtend higher level cognitive phenomena. This approach is of obvious relevance to ITNT. Recent work in evolutionary robotics has precisely replicated classical experiments in animal navigation, allowing rigorous quantitative comparisons between the performance of natural and artificial systems. In the long run it may be possible to re-use this methodology to benchmark neuronally-controlled robots against the performance of experimental animals, as reported in the relevant literature

         

Experimental set up
The strategic goal of Neurovers-IT is to work towards a new class of biologically-based computational device based on networks of dissociated neurons (grown in culture) grown on Multi-Electrode Arrays and   interfaced to conventional electronic equipment for pre- and post-processing of neuronal signals. To verify the possibility of such an approach, the project has developed a novel experimental setup (see figure below) in which dissassociated neural cultures, cultured on on multi-electrode array are coupled in a closed loop to a "Khepera" miniature robot. The neural culture provides output to the robot's motors; the robot sensors provide input to the neural culture as it moves round through the environment. The Neurovers-IT partners believe, that if their underlying hypothesis is correct,   the neural culture will be trainable: i.e. it will be able to learn simple tasks, defined by a researcher.

         
   
       
< prev        
         
         
         
  ©Neurovers-it Consortium 2007
webdesign paololafarina.com