Dynamical perception: Modeling framework, model and experiment

Can a distributed anatomical and functional architecture serve as the basis for sufficiently complex perceptual phenomena? In addressing this question, the conceptual notion of dynamical system and its relation to other paradigms is considered including its definition. The principal goal is to develop a dynamical framework on which to ground the theoretical study of perception and other physical phenomena. As an entry point, the perceptual dynamics of auditory streaming are modeled using a neurally inspired dynamical model of auditory processing. Traditional approaches view streaming as a competition of streams, realized within a tonotopically organized neural network. In contrast, streaming can be viewed as a dynamic integration process involving locations (information convergence zones) other than the sensory specific neural subsystems. This process finds its realization in the synchronization of neural ensembles. Consequently, the model employs two interacting dynamical systems. The first system responds to incoming acoustic stimuli and transforms them into a spatiotemporal neural field dynamics. The second system is a classification system coupled to the neural field and evolves to a stationary state in the absence of input. The states of the classification system at any time t are identified with a single perceptual stream or multiple streams. Several results in human perception are modeled including temporal coherence and fission boundaries (van Noorden, 1975), and crossing of motions (Bregman, 1990). The model predicts phenomena such as the existence of two streams with the same pitch. So far, this has not been explained by the traditional stream competition models. A psychophysical study provides proof of existence of this phenomenon. Using set theoretical expressions on fMRI data, evidence was found showing that the percept of auditory streaming involves regions (convergence zones) other than just the primary auditory areas. This is a necessary condition for the existence of the network architecture proposed in the auditory streaming model. Networks specific and common to both amplitude and frequency streaming were identified. This lends support to models of perception conceived as interacting neural subnetworks acting as functional differentiation areas and information convergence zones for the classification of the perceptual world as suggested by the introductory question.