How the neocortex works is a mystery. Traditional feed forward models of perception cannot account for the vast majority of cortical connections. In this talk, I will describe a theory that sensory regions of the neocortex process two inputs. One input is the well-known sensory data arriving via thalamic relay cells. The second is an allocentric representation of location, which we propose is derived from motion inputs in the sub-granular layers of each cortical column. The allocentric location represents where a sensed feature is relative to the object being sensed. As the sensors move, cortical columns learn complete predictive models of objects by integrating feature and location representations over time. We propose a theory where the location signal is derived in each column using the same principles as grid cells in the entorhinal cortex. In this proposal, individual cortical columns are able to model complete objects and are far more powerful than currently believed. I will discuss our model, the mechanisms, and the implications for hierarchy and cortical function.
About the Centre for Theoretical Neuroscience:
Recently established at the University of Waterloo, the Centre for Theoretical Neuroscience is a growing research initiative comprising multiple labs that focus on theoretical, or computational, neuroscience. The Centre runs a regular academic colloquium series, and Numenta VP Research Subutai Ahmad will be the speaker for the October session.