A Neurodynamical Model of Sequencing and Executing Visual-Mental Operations
1 Institute for Neural Information Processing, Ulm University
To solve demanding perceptual problems, the brain needs to act flexibly utilizing mental representations in a state-dependent manner. Task goals can be achieved by deploying a context-dependent sequence of mental operations. This requires the compositional assembly of operations into routines and a scheduling mechanism that monitors the activation and execution of such individual operations. Here, we propose a neurodynamical model capable of scheduling, executing and monitoring such modular mental operations to form visual routines. The model is grounded in earlier work on competitive queueing to encode event-related patterns of graded neural activity into an ordered serial execution. We extend this model by a population of neurons that provides a means of evidence accumulation and that monitors task progress to steer the sequential execution of neural operations. Furthermore, we provide a dynamical implementation of a previously described synaptic neural pointer idea, which allows flexible indexing of patterns of neural activity from another population. This mechanism helps to form a working memory structure and to abstract away processing content from sequential structure. This enables the serial execution of operations to compose mental routines based on abstract sequence slots, which, in turn, index content-related working memory slots via neural pointers. We exemplify the inner workings of the model on a task of sequential perceptual decision-making based on curve tracing and visual search. We discuss relations to functional brain networks, procedural memory formation, rhythms of processing, and consciousness frameworks.