Spatial Coherence as an Internal Teacher for a Neural Network
The mammalian perceptual system can recognize complex stimuli with remarkable speed and robustness. So far, the performance of computational models of perceptual skills such as the recognition of continuous speech and complex scenes has fallen far short of the performance level of humans on these tasks. How does the brain achieve such remarkable pattern recognition abilities? A partial explanation may be that the brain undergoes substantial self-organization as the young animal gradually learns to find regularities in the complex patterns of stimuli which it encounters. By examining the ways in which the brain employs self-organizing principles, we can gain some insight as to how we should design efficient, and biologically plausible computational models of learning.