Mechanisms for spontaneous symmetry breaking in developing visual cortex

Abstract

For the brain to recognize local orientations within images, neurons must spontaneously break the translation and rotation symmetry of their response functions -- an archetypal example of unsupervised learning. The dominant framework for unsupervised learning in biology is Hebb's principle, but how Hebbian learning could break such symmetries is a longstanding biophysical riddle. Theoretical studies agree that this should require inputs to visual cortex to invert the relative magnitude of their correlations at long distances. Empirical measurements have searched in vain for such an inversion, and report the opposite to be true. We formally approach the question through the hermitianization of a multi-layer model, which maps it into a problem of zero-temperature phase transitions. In the emerging phase diagram, both symmetries break spontaneously as long as (1) recurrent interactions are sufficiently long-range and (2) Hebbian competition is duly accounted for. The relevant mechanism for symmetry breaking is competition among connections sprouting from the same afferent cell. Such competition, and not the structure of the inputs, is capable of triggering the broken-symmetry phase required by image processing. We provide analytic predictions on the relative magnitudes of the relevant length-scales needed for this novel mechanism to occur. These results reconcile experimental observations to the Hebbian paradigm, shed light on a new mechanism for visual cortex development, and contribute to our growing understanding of the relationship between learning and symmetry breaking.