Improving Spatiotopic Models of Vision using Retinotopic Input and Vector-based Saccadic Generation: A theoretical and methodological framework for the study of attentional control settings

Abstract

Our surroundings are seldom stable and are filled with various information. Visual processing is one of the ways the brain deals with all this information; however, the capacity of our visual system is limited. To deal with this limitation, we have attentional mechanisms that help us extract relevant information from noisy surroundings. Within the visual modality, we can study these attentional mechanisms through the measurement of eye movements and computational approaches. This dissertation proposes a theoretical framework for the potential combination of eye movement data and computational approaches to construct a biologically inspired computational model of visual attention that would accommodate the attentional control settings under various context and task conditions. Machine learning has gained popularity in the recent years. Many sophisticated algorithms and approaches fuel technological developments, such as face and object recognition for robotic applications like self-driving cars or security software. While these areas of research and technological application are undoubtedly important and interesting, there remain other domains that could benefit from such approaches, such as vision modelling. This dissertation aims to address existing gaps in the modelling of visual attention within the domain of cognitive neuroscience vision research. Specifically, the aims are: (1) to analyse the state-of-the-art in the field of visual attention modelling with respect to computational cognitive neuroscience; (2) to gain a deeper insight into some of the cognitive and biological mechanisms in visual attention; and (3) to introduce ways to improve existing computational models of vision using the results of the studies. The aims are addressed in three studies. Study I focuses on computational modelling and consists of one major and two auxiliary manuscripts. The former provides an in-depth review of the field and proposes a novel theoretical guideline for computational cognitive neuroscience models of visual salience. The latter two papers investigate some computational approaches to create a biologically accurate model of visual attention, as well as their limitations. In Study II attentional control settings are observed within a gaze-contingent experimental setup, where the size of observers’ functional field of view is directly manipulated. Performance (response times and accuracy rates) is assessed across several task conditions, revealing that the size of the functional field of view is dependent on more factors than initially thought. Study III proposes a novel experimental design to studying executive control and oculomotor suppression via microsaccade rates in the antisaccade paradigm. The results of this study demonstrate that microsaccade rates can be used as a measure of the degree of suppression of the oculomotor system. The theoretical and experimental findings of the studies included in this dissertation are important for the field of cognitive computational modelling of visual attention and perception. The insights gained from them can improve our understanding of how visual attention operates within different contexts, while the use of computational modelling makes it possible to observe the interactions of the attentional mechanisms that guide our perception of the surroundings.