COMPUTATIONAL ARCHITECTURE OF HIGH-LEVEL ATTENTION- REVERSE-ENGINEERING REPRESENTATIONS AND GOALS THAT DRIVE SEEING IN COMPLEX, DYNAMIC ENVIRONMENTS
Abstract
Real-world environments are complex and dynamic, consisting of many objects, surfaces, agents, and their multiplied interactions. Yet, given a goal or a subtask, only a few entities in the environment matter and they matter to varying degrees. Perceptual representations (e.g., objects, scenes, agents, and events) are under attentional control, selectively and dynamically deployed in a goal-driven, flexible manner. Despite its centrality and ever-presence, none of the existing approaches, since the beginning of cognitive revolution, were able to address the computational nature of goal-driven mental processes in perception. A new conceptual framework for understanding perception and attention that goes beyond existing approaches is needed to address how and why we selectively process objects, scenes, events, and agents. This proposal addresses this knowledge gap to understand representations, implicit goals, and algorithmic motifs that drive visual cognition under complex, dynamic environments, in an integrative program linking phenomena at the cognitive and neural levels, by building a computational architecture of high-level attention. Our overarching hypothesis is that high-level attention can be understood as implicit (or explicit, if present) goals of an observer driving the formation of a percept through rational allocation of computing cycles and representational resources. This account of high-level attention opens a new, revealing window into visual cognition, through which, in static and dynamic scene perception, we will query phenomena that so far have largely eluded modeling.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210041
Entities
People
- Ilker Yildirim
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- Yale University