Cognitive Programs Memory - A Framework for Integrating Executive Control in STAR

Cognitive Programs are algorithms which guide the execution of visual tasks in the Selective Tuning (ST) model and its extension Selective Tuning Attentive Reference (STAR). However, no framework exists that allows for their rapid development, parameterization and execution. In this thesis we propose that a basis set of elemental operations called Neural Primitives (NP) in conjunction with other control elements constitute a Cognitive Program. The Neural Primitives are biologically inspired computations that dictate the transformation functions from one representation to another and form the foundation of this thesis. Therefore a key theme here is using low level neural computations, the NPs to perform higher level cognitive functions that are required for task demands. This thesis also introduces a database like structure called the Cognitive Programs Memory (CPM) that holds a set of predefined CPs which are accessible by the visual task executive (vTE) - a component of STAR with a major role of controlling task execution. The CPs have been tested on the execution of three psychophysical experiments demonstrat- ing how a sequence of operations can be used to complete visual tasks. The CPs have also been tested on two experiments demonstrating that the application of the NP operations result in qualitatively similar neural tuning curves to neurophysiological data.