Basic of Pattern Recognition
by Matthew C. Putman
The human mind has abilities which far exceed anything that modern computation can achieve. This is most evident in pattern recognition, which includes most sensory inputs. It is only humans and other animals that can truly fill in missing sensory information to such a strong degree as to recognize characters, sounds, faces and more in nearly all locations and times. Much of modern machine learning has been dedicated to computationally replicating the power of the brain in order to eventually exceed it. Ray Kurzweil details some current approaches to the problem in his 2012 book “How to Create a Mind”. In the book he discusses the now applied use of big data and statistical modeling to provide recognition of sensory input. This is an attempt to gain the necessary information to mimic the cortical hierarchy that makes humans so powerful in this regard. The success of this type of approach, which uses probability theory such as Bayesian logic and Hidden Markov Models, can be seen in everyday products that are made by Google and others. This is especially apparent in the facial recognition Google uses with Picassa and in Google Translate. Both of these rely on the very large quantities of data generated by users to gain a human-like memory. An even more impressive example of machine success is the IBM computer Watson, which successfully used learning techniques including the above mentioned modeling, and powerful processing to defeat the world’s best Jeopardy players.