Review of Zoccolan, Cox, DiCarlo (2005) Multiple object response normalization in monkey inferotemporal cortex.

Even though the neurons in inferotemporal cortex (IT) have very large receptive fields, it is tempting the believe that the neurons would be able to distinguish objects presented within their receptive fields. For example, if a neuron responds to object A and B at different rates, perhaps the neuron should give the maximum of these two rates when both stimuli are presented within their receptive field. The study (Zoccolan, Cox, and DiCarlo 2005) shows that this is not the case and, when presented with two objects, most IT neurons’ responses are the mean of the firing rates when the objects are presented separately - at least for short presentation times and when the objects are not attended.

There is a lot more to this paper than what I will cover in this review / note. I hope to add more in the future, but the most important points are straightforward. They use simple artificial shapes on a plain background. The first results show that in the population, the cells’ responses to the presentation of multiple objects cluster around the mean of their responses of when the objects are presented separately. There is slight tendency to fire at a rate slightly higher than the average, but the lack of scatter is rather amazing. There is a line in Figure 1C and 1D for the sum responses and very few of the cells fall on or above this line.

They then show that the responses to the combined object displays are much more like the mean of the responses to individual object displays than a max model, at least in the mean cell population. There is a lot of spread in these results, leaving open the possibility that some neurons give a response that is the maximum of the response to the two objects separately (or having an even higher response).

Zoccolan, Davide, David D. Cox, and James J. DiCarlo. 2005. “Multiple Object Response Normalization in Monkey Inferotemporal Cortex.” J. Neurosci. 25 (36): 8150–64.