INFEROTEMPORAL UNITS IN SELECTIVE VISUAL-ATTENTION AND SHORT-TERM-MEMORY

This research was designed to further clarify how, in the primate, the neurons of the inferotemporal (IT) cortex support the cognitive functions of visually guided behavior. Specifically, the aim was to determine the role of those neurons in 1) selective attention to behaviorally relevant features of the visual environment and 2) retention of those features in temporary memory. Monkeys were trained in a memory task in which they had to discriminate and retain individual features of compound stimuli, each stimulus consisting of a colored disk with a gray symbol in the middle. A trial began with brief presentation of one such stimulus, the sample for the trial. Depending on the symbol in it, the monkey had to memorize the symbol itself or the background color; after 10-20 s of delay (retention period), two compound stimuli appeared, and the animal had to choose the one with the symbol or with the color of the sample. Thus the test required attention to the symbol, in some trials also to the color, and short-term retention of the distinctive feature for each trial, either a symbol or a color. Single-unit activity was recorded from cortex of the IT convexity, lower and upper banks of the superior temporal sulcus (STS), and from striate cortex (V1). Firing frequency was analyzed during intertrial periods and during the entirety of every trial, except for the (match) choice period. In IT cortex, as in V1, many units responded to the sample stimulus. Some responded indiscriminately to all samples, whereas others responded selectively to one of their features, i.e., to one symbol or to one color. Fifteen percent of the IT units were symbol selective and 21% color selective. These neurons appeared capable of extracting individual features from complex stimuli. Some color cells (color-attentive units) responded significantly more to their preferred color when it was relevant (i.e., had to be retained) than when it was not. The latency of IT-unit response to the sample stimulus was, on the average, relatively short in unselective units (mean 159 ms), longer in symbol units (mean 203 ms), and longest in color-attentive units (mean 270 ms). This order of latencies corresponds to the presumed order of participation of those three types of units in the selective attention to the component features of the sample as required by the task. It suggests intervening steps of serial processing before color information reached color-attentive cells. The magnitude of the enhancing effect of relevance on the response of color-selective cells to the sample was found directly related to their response latency. This relationship suggests that selective attention to color resulted in the sequential recruitment of IT cells progressively more susceptible to the behavioral relevance of color. Some units showed sustained firing elevations during the delay. In 14% of all IT units, those elevations depended on a particular feature of the sample, the symbol or the color. The feature selectivity of an IT cell during the delay was not necessarily the continuation of similar selectivity during the sample period: certain cells that did not respond selectively to a feature while the sample was present showed feature-selective firing during the delay, after the sample had disappeared. In conclusion, a number of IT neurons seem to participate in the short-term retention of visual features for behavioral use. Some transfer of information may occur from neurons that encode a feature when present in the visual field to neurons that retain it afterward in short-term memory.