NEURAL PATTERNS AS DETECTION UNITS IN SENSORY SYSTEMS.

A mathematical model describing the first stages of information processing in the visual system is defined and used as a basis for computer simulation for studying the complex patterns of neural activity at various levels of the hierarchy of visual information processing. This simulation features a friendly human interface, produces good graphical displays of the responses of the neuron networks, and requires a reasonable amount of computer time. The model describes the behavior of the receptive fields of X-retinal, X-lateral geniculate nucleus (LGN), and primary cortical cells. The receptive fields of the retinal ganglion or LGN cells are based on the difference-of-Gaussians (DOG) model. This study is an attempt to understand the rules governing transitions between single-cell detectors and neuronal pattern representations in sensory systems. Primary cortical neurons are shown to be the first in the visual system that may be considered as feature detectors. Their interactions with lower-order neurons, their X-cell inputs, and with higher-order neurons may be typical of representation transitions.