Computational and neural models of oculomotor control.

• Main Author: Wilming, Niklas
• Other Authors: Prof. Dr. Peter König
• Format: Doctoral Thesis
• Language: English
• Published: 2015
• Subjects: Aufmerksamkeit; Augenbewegungen; Computermodellierung; fMRI; Attention; Eye Tracking; Computational modelling; ddc:500; ddc:150
• Online Access: https://repositorium.ub.uni-osnabrueck.de/handle/urn:nbn:de:gbv:700-2015030913109

Seeing is more than sight: it is the entire action-perception loop involved in taking in the world around us. Unlike a camera, our eyes can only resolve a small part of the environment sharply. Therefore, we must constantly move our eyes to scrutinise the parts of our environment that seem most worthy of our highest visual acuity. Eye movements are thus the observable consequences of a complex and crucial decision-making process that is fundamental to how we interact with the world. This thesis investigates properties and the neural basis of eye-movement behavior in humans and monkeys. In the interdisciplinary tradition of cognitive science, the thesis spans fields and utilizes computational models as explanatory vehicles. A central theme is the so-called saliency map model of attention, the de facto computational model of viewing behavior. The saliency map model assumes that attention is directed at the peaks of a map that encodes the saliency of locations in the visual field. Saliency can roughly be thought of as how worthy a location is of attention. It forms a common currency that allows different processes to influence the distribution of attention. The four different studies in this thesis provide four different perspectives on viewing behavior and the saliency map model. The first study establishes a methodology to evaluate the predictive power of models of viewing behavior, and determines which properties of viewing behavior are important for this evaluation. Applying this methodological foundation to the saliency map model reveals that state-of-the-art models do not provide satisfactory explanations of viewing behavior. The second study investigates spatio-temporal properties of eye-movements, finding that observers often re-fixate locations in pictures and that their eye movements possess a rich spatio-temporal structure. These results speak directly against a causal role of "inhibition of return", which is a popular component of many saliency map models. The third study shifts focus to the neural basis of the oculomotor behaviour. fMRI is used to probe the relationship between the computation of saliency and actual processing in the brain. Our results, in contrast to those of other studies, suggest that early visual areas do not compute saliency, but instead compute visual features upon which the saliency map operates. Much of what we know about the neural basis of oculomotor control comes from invasive studies in animals, but it is unclear to what extent saliency computations are comparable between species. Thus, the fourth study compares the viewing behavior of monkeys and humans, to look for evidence of the same underlying processes. We find a strong similarity between the species in saliency-driven viewing behavior. The many saliency-processing areas that have been identified in monkeys therefore likely have a role in saliency processing in the human brain as well. This thesis contributes to our understanding of oculomotor control on multiple levels. The results in this thesis suggest that models of viewing behavior should treat saccade-target selection as a dynamic process where past decisions influence future decisions and where saliency varies over time. This selection process likely takes place in a distributed network in the brain which receives bottom-up input from early visual areas. Encouraged by these results, we speculate that normative and embodied models of cognition offer an explanation of oculomotor control that takes these results into account. In turn, explaining oculomotor control is an important part of the much deeper question of how our mind interacts with the world.