Attention in model-based cognitive neuroscience: Mechanisms for optimization under constraint

Two major criticisms of attention research are: 1) verbal theories lack the specificity to make falsifiable predictions about how the various mechanisms of attention operate; and 2) the term ``attention’’ is frequently ascribed to conceptually similar phenomena despite dissociable neural underpinnings. The field of model-based cognitive neuroscience is well-positioned to address (1) through mathematical formalization of hypothesized processes, but may perpetuate (2) insofar as similar algorithms and parameters may recur across modeling traditions without necessarily drawing upon a shared neuro-mechanistic basis. This review provides a function-first organization of the algorithms that have been hypothesized to implement or give rise to ``attention'' at various levels of processing: perception, decision-making, and learning. We ground mathematically-formalized processes in the performance-relevant objectives they serve and the processing constraints that limit them. In doing so, we provide a road map of the broader computational-level solutions that ``attention'' mechanisms have been invoked to provide.