Solving the Sensory Information Bottleneck to Central Processing in Adaptive Systems

Abstract

Information bottlenecks are an inevitable consequence when a complex system adapts by increasing its information input. Input and output bottlenecks are due to geometrical limits that arise because the area available for connections from an external surface always exceeds the area available for connections to an internal surface. Processing of the additional information faces an internal bottleneck As more elements increase the size of a processor, its interface surface increases more slowly than its volume. These bottlenecks had to be overcome before more complex life forms could evolve. Based on mapping studies, it is generally agreed that sensory inputs to the brain are organized as convergent-divergent networks. However, no one has previously explained how such networks can conserve the location and magnitude of any particular stimulus. The solution to a convergent-divergent network that overcomes bottleneck problems turns out to be surprisingly simple and yet restricted.