Abstract

D. J. Berg, S. E. Boehnke, R. A. Marino, P. F. Baldi, D. P. Munoz, L. Itti, Monkeys as a Model for Human Bottom-Up Overt Attention, In: Collaborative Research in Computational Neuroscience Annual Meeting, Los Angeles, California, Jun 2008. (Cited by 3)

Abstract: Monkeys are widely used as animal models for human visual attention, however, they are usually studied using simplified stimulus conditions in order to have experimental control for neural recording. Here we compared the visual behavior of 5 human and 5 monkey observers to determine the degree that the two species show bottom-up differences in visual behavior. To this end we recording eye movements of naïve observers in a free-viewing task using 115 video clips (47,903 frames, approximately 27 minutes) ranging in semantic content (natural scenes, monkey relevant, noise etc). Saccade and fixation statistics such as the main sequence, saccadic endpoint distribution and frequency of saccades indicate a strong difference between human and monkey eye movements. To test whether these difference affect overall visual behavior and to establish to what extent human and rhesus monkey eye movements can be predicted by bottom-up stimulus properties we quantified primate gaze shifts using a model of visual attention, Surprise Model (Itti and Baldi '06). We also compared a given subjects gaze to the gaze locations of all other subjects on the same stimulus(Interobserver model), which includes bottom-up and top-down factors. A computational model of visual attention predicted human gaze shifts marginally, but significantly better than monkey gaze shifts(p < 0.001). However, when looking at individual video clips the correlation between measured Surprise at human and monkey saccadic endpoints is high (r2 = 0.62, p < 10-9). Furthermore, locations that 3 monkeys looked at simultaneously were higher in Surprise than those locations that multiple humans agreed to look (p < 0.07). The computational model predicted monkey eye movements better than the monkey interobserver model (p < 0.002), yet the interobserver model predicted human gaze shifts far better than the computational model (p < .001). In whole the results demonstrate that although there are some differences in visual behavior between species, monkeys can be effectively used as a model for stimulus driven attention in humans during free viewing tasks.

Themes: Model of Top-Down Attentional Modulation, Computational Modeling, Human Psychophysics