training neural nets from real signals

Jean-Michel FRIEDT jmfriedt at lpmorh04.univ-fcomte.fr
Fri Dec 4 03:08:02 EST 1998


I am wondering if this project I am thinking of has already been
achieved, and if not why. First of all, I am not a biologist but
physicist/programmer, so excuse the naive questions on biology.

I have been thinking of connecting the output of electrodes
connected to the ganglions of a cockroach to a neural net and
train the neural network with these signals. 

The training of the neural net would involve comparing the signal
read by the electrodes to the behaviour of the cockroach. I would expect,
with enough samples, the neural net to be able to extract the 
signal I am looking for (let's say, motion of one of the antenna) from
all the signals available at the ganglion level. So the setup would
be something like that :
  ganglion --> electrode --> data acquisition <-- antenna position sensor
				     |
                               neural network

I am thinking of asking a neurobiology team to train me on the biological
part of this idea (I am PhD student in scientific intrumentation right now),
considering that the data acquisition, position sensor and neural net are
easy for me to develop.

So, has this already been done (I couldn't find any bibliography on the
subject other than Prof. Miura's work which is not exactly what I want to
do), and if not why ? 

Any help would be appreciated, thank you for your time,
   Jean-Michel FRIEDT
   jmfriedt at lpmorh04.univ-fcomte.fr
  (jmfriedt at ens-lyon.fr)




More information about the Neur-sci mailing list