Neurons in Primate Visual Cortex Alternate between Responses to Multiple Stimuli in Their Receptive Field

Abstract

A fundamental question concerning representation of the visual world in our brain is how a cortical cell responds when presented with more than a single stimulus. We find supportive evidence that most cells presented with a pair of stimuli respond predominantly to one stimulus at a time, rather than a weighted average response. Traditionally, the firing rate is assumed to be a weighted average of the firing rates to the individual stimuli (response-averaging model) (Bundesen et al., 2005). Here, we also evaluate a probability-mixing model (Bundesen et al., 2005), where neurons temporally multiplex the responses to the individual stimuli. This provides a mechanism by which the representational identity of multiple stimuli in complex visual scenes can be maintained despite the large receptive fields in higher extrastriate visual cortex in primates. We compare the two models through analysis of data from single cells in the middle temporal visual area (MT) of rhesus monkeys when presented with two separate stimuli inside their receptive field with attention directed to one of the two stimuli or outside the receptive field. The spike trains were modeled by stochastic point processes, including memory effects of past spikes and attentional effects, and statistical model selection between the two models was performed by information theoretic measures as well as the predictive accuracy of the models. As an auxiliary measure, we also tested for uni- or multimodality in interspike interval distributions, and performed a correlation analysis of simultaneously recorded pairs of neurons, to evaluate population behavior.

Sensitivity and specificity of a Bayesian single trial analysis for time varying neural signals

“…It is well known that such firing patterns are variable in the face of identical, highly controlled experimental conditions (such as the presentation of the same stimulus in the same context). While many studies have viewed this variability as deleterious "noise" that if unsolved would undermine the ability of the brain to perform its essential tasks [12][13][14][15][16] , we and others have sought explanations under the possibility that certain forms of such variation may contribute in a positive fashion to brain function 1,[17][18][19][20][21][22][23][24] . In particular, we have successfully modeled variation in whole-trial spike counts to multiple stimuli as being drawn from the observed distributions of spike counts to those same stimuli when presented individually 1 .…”

Distinguishing between parallel and serial processing in visual attention from neurobiological data

“…Furthermore, we propose measures to quantify the processing 69 mechanism in a continuum between serial and parallel processing. The method relies on 70 the probability-mixing model for single neuron processing [16], derived from the Neural 71 Theory of Visual Attention [1,17], which states that when presented with a plurality of 72 stimuli a neuron only responds to one stimulus at any given time. By probabilistic 73 modeling and statistical inference using multiple simultaneously recorded spike trains, 74 we infer and decode what the recorded neurons are responding to, providing a way to 75 distinguish between parallel processing and serial processing on a neuronal level.…”

“…Consider an experiment in which we record the action potentials or spikes from each 79 of a number of neurons of the same type, e.g., a set of functionally similar neurons in 80 visual cortex with overlapping receptive fields (see, e.g., [16]), or neurons in prefrontal 81 cortex that are believed to be dynamically allocated to process task-relevant 82 information (see, e.g., [4]), which are the neurons we analyze in this paper. Suppose two 83 stimuli (Stimulus 1 and 2) are both within the classical receptive fields of all of the 84 recorded neurons, but otherwise the receptive fields are empty.…”

“…195 We assume a homogeneous situation where all neurons follow the same distribution and 196 are exchangeable, except for individual firing rates as responses to single stimuli. These 197 measures are based on the basic probability-mixing model for the attention of single 198 neurons employed in [16], where a neuron responds to a stimulus mixture with certain 199 probabilities, such that the single neuron at any given time represents only one of the 200 stimuli in the mixture. First, we consider the marginal distribution of the attended 201 stimulus for each neuron.…”

“…The 309 asymptotic version of the deviation statistic D * to measure the degree of serial and 310 parallel processing is given by parameter settings, shown in Table 3, together with the D 10 and D * values. The spike trains are modelled by point processes using conditional intensity functions (CIF) [20,21], see also [16]. Suppose a spike train y in the interval [T s , T e ] contains the spike times y = {t 1 , t 2 , .…”