Hi Sai, The CLA represents a very tiny slice of the neocortex, maybe 1mm squared (compared with perhaps 50-1000mm squared) of one layer (out of 5 or 6 highly connected layers) of one region (which normally exists in a big hierarchy of dozens of regions). So it won't do facial recognition out of the box!
Facial recognition is a hierarchical problem, so you need to have a hierarchical solution. You can do this with the CLA if you can feed it sufficiently predigested data, for example: This face is w in width, h in height; nose length is n1, bulbousness is n2, brokenness is n3; eyes are y1 apart, y2 in size, colour y3, slantiness y4, eyelashes are y5 long, y6 in thickness; ears are e1 below eyeline, e2 in size, e3 in earlobe size; mouth is m1 below eyeline, m2 in width, m3 in fullness... Humans get this kind of data by using lower-level regions to extract it, mid-level regions to identify each measurement, and higher-level regions to fit them together into a learned pattern for facial recognition. Each region in each layer is a CLA process, that's the theory. So, to do this you can predigest the data using some existing vision algorithms (currently in use for analysing CCTV pictures, or Facebook photos) and feed to a single CLA, or you can build a hierarchy of CLAs to do everything. Regards, Fergal Byrne
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
