Model study on the spontaneous emergence of functional organizations in the visual system

Abstract

Understanding the developmental mechanism of functional organizations in the visual pathway has provided insight into the emergence of intelligence in the brain. For example, as the functional unit processing the basic visual function, neurons in the primary visual cortex (V1) are tuned to the orientation of visual stimuli, and the preferred orientation of neurons is arranged quasi-periodically across the cortical surface, termed an orientation map. Previously, most studies on the emergence of these functional organizations have focused on the intra-cortical developmental mechanism modulated by the postnatal cortical activity triggered by visual experience. However, several functional organizations have been observed in early developmental stages even before the visual experience, implying that the emergence of these organizations may be initiated by a more fundamental mechanism that can operate without any visual experience. Herein, using a computer simulation of neural network, I show that the emergence of functional organizations could be spontaneously initiated by the haphazard wiring of feedforward projections, even in the absence of any further refining process. The early part shows that the structure of ON and OFF retinal ganglion cell (RGC) mosaics is projected onto V1 by the retino-cortical feedforward mapping. This projection initiates the emergence of three main characteristics of functional maps in V1, including the consistency of spatial periods across animals seeded by an interference pattern of ON and OFF RGC mosaics having a similar period, the development of disparate organizations across species depending on their conditions of retino-cortical mapping, and the systematic organization of diverse functional maps initiated by a common structural source from retinal mosaics. In the later part, the simulation using artificial deep neural networks as a model of further visual pathway shows that a higher cognitive function, in this case number sense, could also be initiated by the feedforward projections of the randomly-initialized network, even in the complete absence of learning. In summary, these results suggest that the emergence of functional circuits processing cognitive functions is solely initiated by the physical structure of feedforward neural networks, shedding light on the origin of the spontaneous emergence of the intelligence.