Unsupervised Learning through Prediction in a Model of Cortex

Human infants can do some amazing things, and so can computers, but there seems to be almost no intersection or direct connection between these two spheres of accomplishment. In Computer Science we model computation through algorithms and running times, but such modeling quickly leads to intractability, even when applied to tasks that are very easy for humans. The algorithms we invent are clever, complex and sophisticated, and yet they work in fashions that seem completely incompatible with our understanding of the ways in which the brain must actually work -- and this includes learning algorithms. Accelerating advances in neuroscience have expanded tremendously our understanding of the brain, its neurons and their synapses, mechanisms, and connections, and yet no overarching theory appears to be emerging of brain function and the genesis of the mind. As far as we know, and the spectacular successes of neural networks [5, 8] notwithstanding, no algorithm has been proposed which solves some nontrivial computational problem in a computational fashion and style that can be credibly claimed to reflect what is happening in the brain when the same problem is solved.