Title: Neural network models of visual search and adaptive saccadic eye movements
Abstract: Saccades are rapid eye movements which allow an animal to direct a limited resource, the fovea, toward behaviorally relevant portions of the world. Control of saccades involves at least two tasks. First, a target is selected from the visual scene. Second, the eyes are moved to direct the fovea accurately at the selected target. This dissertation models aspects of how these two tasks are accomplished in the brain.
Targets may be selected by spatial attention prior to movement of the eyes. The first part of this dissertation develops a model of attentive visual search in which feature stages, which analyze a scene in parallel, guide a serial stage, which processes a single item at a time. In the model, attention interacts with object recognition and the programming of saccadic eye movements. Model simulations reproduce visual search set size and target eccentricity effects, as well as saccadic averaging and scanpaths.
Saccades can also be made reactively to visual input or in response to pre-planned locations. These various saccade types are mediated through different brain regions. They maintain a balance between the ability to react rapidly and the ability to perform complex, planned behaviors. When saccadic commands differ, the superior colliculus helps to select which brain area gains control of the eye muscles. Different saccade types may lead to different patterns of superior colliculus activation. Learning is required to produce accurate saccades in all cases.
A neural model is developed of how these brain regions interact with the saccadic system and use cerebellar gain learning and cortical map learning to maintain saccadic accuracy during reactive, attentive, and planned saccades. These saccade types are mediated by separate processing streams in the model. Model simulations explain task-specific adaptation data concerning learning transfer or lack thereof between different saccade types. Model adaptation transfer depends on saccade latency, as in experimental data. The model makes testable predictions about adaptation transfer in cases that have not yet been experimentally studied. The saccade generator portion of the model produces saccadic staircases, accurate interrupted saccades, straight oblique saccades, velocity duration tradeoffs, and smooth eye movements at high input levels.
Publication Year: 1998
Publication Date: 1998-01-01
Language: en
Type: article
Access and Citation
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot