Joyce Tam¹ (), Chloe Callahan-Flintoft², Brad Wyble¹; ¹Penn State University, ²US Army DEVCOM Army Research Laboratory

A recent set of studies reported that target-directed saccades are not necessarily preceded by an N2pc, casting doubts on the obligatory relationship between covert selection and oculomotor programming. Here, we present RAGNAROC-PLUS (RAG+), a computational model that addresses this uncertain relationship between saccadic decisions and the N2pc while also explaining other well-established empirical benchmarks in visual search. RAG+ is an integrative theory of covert and overt attentional control through the coordination of ventral cortical and subcortical areas. The model simulates both covert and overt search metrics including event-related potentials, saccadic accuracy and latency, and behavioral performance in a variety of search paradigms. Specifically, the model is composed of multiple sheets of computational neuron maps, organized in a hierarchical and pyramidal manner, representing visual information in a few different spatial resolutions. Signals travel in both feedforward (analogous to posterior-to-anterior) and feedback directions, wherein neuron activations are modulated by external information (visual saliency and task relevance) and undergo a series of mutual excitatory and inhibitory interactions, forming a hierarchical attentional landscape that evolves over time. Importantly, while the N2pc is exclusively linked to feedback neural activities, as per previous findings, here we show how highly accurate saccadic decisions can be made rapidly with feedforward-only attention (i.e., no N2pc), which is the case when the level of spatial ambiguity in the display is low. Through this approach, we show how metrics of covert and overt attention can be generated within highly overlapping structures and yet do not always co-occur. RAG+ showcases the use of computational modeling to identify specificities that differentiate intimately related concepts such as covert and overt attention. Moreover, our model represents a step forward in the understanding of attentional dynamics in the real world where covert and overt attention often work in conjunction.

Acknowledgements: This project is supported by National Science Foundation grant awarded to B. W. (NO. 1734220) and US Army DEVCOM Army Research Laboratory Research Associateship Program Journeyman Fellowship awarded to J. T. (CA: W911NF-22-2-0097).