From basic network principles to neural architecture: Emergence of orientation-selective cells

Abstract

This is the second paper in a series of three that explores the emergence of several prominent features of the functional architecture of visual cortex, in a 'modular self-adaptive network' containing several layers of cells with parallel feedforward connections whose strengths develop according to a Hebb-type correlation-rewarding rule. In the present paper I show that orientation-selective cells, similar to the 'simple' cortical cells of Hubel and Wiesel, emerge in such a network. No orientation preference is specified to the system at any stage, the orientation-selective cell layer emerges even in the absence of environmental input to the system, and none of the basic developmental rules is specific to visual processing.