Using EEG and Neural Network Modeling to Determine the Dynamics of Visual Selection - Project Summary The proposed work aims to identify the neural mechanisms of visual selection, which is heightened visual salience of task-relevant information. The visual system receives constant sensory input that must be parsed to select important information. For instance, drivers must detect road signs, cars, and pedestrians to determine the proper response. In the lab, this is studied using Rapid Serial Visual Presentation (RSVP) in which stimuli are presented at 10 Hz in one location and observers must respond to pre-defined “targets” (e.g., report letters, but not digits). Although the visual system can select one target at this rate, it fails to select a second presented shortly after. This transient deficit has been termed an “attentional blink”; however, the research is a mixture of single-task (e.g., report letters, not digits) and task-switching paradigms (e.g., report a white letter, then a black X). Indeed, task-switching variants may be due to an attentional deficit, single-task variants potentially reflect visual dynamics. In brief, the single-task second target deficit might reflect “conceptual repetition blindness”, defined as a transient inability to perceive that the second target belongs to the category of targets. I report data from two pilot experiments in which observers report either two words defined by their category (single-task) or one word representing a number and another defined by category (task-switching). Different patterns emerged which suggests the involvement of separate neural mechanisms. Prior work is mixed regarding whether N400 components track the second target deficit with a key difference being whether the tasks are switching. To properly test this hypothesis, I plan to conduct similar experiments utilizing semantic word stimuli and test whether known markers of early semantic processing (N170 ERP component) and memory updating (N400) are differently influenced based on task. More specifically, I predict that early, visual ERPs will also track the second target deficit in the single-task version but not in the task-switching version. This hypothesis will be tested across three aims that assess different kinds of visual selection. Aim 1 will examine semantic selection. Aim 2 will examine orthographic case selection (e.g., upper versus lower case), and Aim 3 will examine objects class (e.g., words versus faces). In each aim, it is predicted that the single-task version will reveal a visual ERP component that tracks the second target deficit, but the nature of the component will depend on the nature of the selection. For instance, orthographic case selection might reflect an earlier component (e.g., the P100), whereas object class (words versus faces), might reflect different hemispheres. I will use machine learning classifiers to identify EEG responses that best predict behavior and stimulus class. Finally, Aim 4 will test this theory by applying a dynamic neural network model that previously explained repetition blindness as the consequence of neural habituation. Using equivalent dipole modeling, the model will be applied to the full RSVP sequence to identify the cortical locations of the layers underlying the second target deficit.
