Perceptual grouping in a self-organizing map of spiking neurons
Abstract
Perceptual grouping is the process of identifying the constituents in the visual
scene that together form a coherent object. The goal of this thesis is to understand
the neural mechanisms of perceptual grouping. The hypotheses are that
(1) perceptual grouping is carried out through synchronized firing of neurons representing
the same object, and that (2) self-organized lateral connections encoding
statistical regularities of the visual environment mediate such a synchronization. A
self-organizing neural network of spiking neurons was developed to test these hypotheses
in the perceptual grouping task of contour integration. The network selforganized
orientation maps and patchy lateral connections similar to those found in
the visual cortex, and the contour integration, segmentation, and completion performance
measured by the degree of synchrony in neural populations accurately predicted
human performance. Such results suggest that synchronized activity can represent perceptual events, and statistical properties of the input can shape the structure
of the cortex and the perceptual performance. By providing a computational
framework where perceptual performance and neural structure can be compared,
the model helps us understand the neural mechanisms of perceptual grouping.
Department
Description
text