CNS 246: Multiunit Recording 
Project Proposal
Evaluation of Noise Correlation in Simultaneously Recorded Neurons

Alejandro Backer
October 20, 1996

Neuronal responses to successive, spaced, presentations of any given 
stimulus vary from trial to trial. This variation has traditionally been 
represented as the result of noise superimposed on the stimulus signal. 
Alternatively, the variation could represent the multiplexing of other 
signals not directly related to the experimenters' stimulus. 

Of course, one could take the view that if a group of cells are to 
represent a certain stimulus parameter, then any variation in response 
due to other signals is just noise with respect to the represented 
signal. The only valid distinction, I believe, is whether the other 
signals are read by downstream neurons or not. If variation is not used 
(read) by the brain, we can call it noise for the process of extracting 
the signals the brain does use. If the variation is used, then noise is 
just a euphemism for our ignorance.

Another related issue is whether the noise, if the inter-trial variation 
can be called noise, is correlated among different neurons. This has 
implications for the possibility of averaging out such noise by pooling 
the responses of a population of neurons, and the number of neurons that 
could be usefully included in such a pooling population.

This correlation could also tell us something about whether the 
variability carries information or not. If the noise was independent from 
neuron to neuron, it would appear most likely that it was noise, 
independently generated in each neuron. If, instead, the variability was 
due to some factor related to the state of the animal or to some 
uncontrolled stimulus property, then one would expect that a deviation 
from the average response of one cell would be accompanied by a deviation 
from the average response (averaged over trials) of another cell.

Recordings simultaneously obtained by Mike Wehr from pairs (and some 
triplets) of neurons of the  antennal olfactory lobe of the locust during 
olfactory stimulation provide us with the possibility of addressing these 
questions directly. Any one of these cells shows markedly similar 
responses over repeated presentations of any given odor. These responses 
are different from cell to cell, and from odor to odor. However, the 
consistency of response for different trials of the same odor is not 
absolute: there is some variability, which has been characterized by 
saying that the neurons exhibit given probabilities of spiking during 
each of the cycles of a periodic population activity that occurs 
transiently in response to odor presentation. 

I plan to rate each spike train (each neuron, each trial) with respect to 
how much it deviated from the typical spike train, as defined by the 
PSTH. The rating of each spike train would be given by the 
multiplication, for all time bins, of the probability of that neuron 
firing in that time bin(averaged over all trials for that neuron) for all 
the time bins in which a spike occurred in that spike train, times the 
trial-averaged probability of the neuron not firing during that time bin 
for all time bins in which a spike did not occur. Thus, spike trains 
resembling the average spike train will get a larger rating than those 
differing substantially.

I will then correlate the "typicality" ratings for simultaneously 
recorded cells as a function of trial, i.e. I will test whether atypical 
response in one cell is accompanied beyond chance-level by atypical 
response of another cell during that same trial.

This will assess the independence of what is currently considered to be 
noise in the neural response of these neurons, establish the degree of 
noise correlation, and possibly illuminate us with regard to the noise 
versus information question of neuronal response variability.