Pattern Memory Inc.
(About Us)
Software for Machine Learning
Home Technology Applications Products About Us Contact Us
Software for Machine Learning
Home Technology Applications Products About Us Contact Us
From the President and Inventor of PME
My personal interest in pattern recognition began 30 years ago as an
Engineering and Biophysics graduate student at the University of
Minnesota. My passionate interest was: "Why can't computers
learn to recognize visual patterns like we
do"? The answer was obvious at that time, computers were slow, cameras
had low resolution, and memory capacity was limited. Thirty years later with
many orders of magnitudes performance increase in all aspects related to
computers - still not much progress compared to what we do naturally
(e.g. visual recognition and continuous speech understanding).
The current barriers relate to the use of
complex mathematics to represent and process
patterns. We can be sure that there is nothing
resembling a computer doing mathematical
computations in biological systems. Also, we learn
patterns after many examples from simple building
blocks to more complex patterns. Neuroscientists
generally agree that biological systems represent
and learn patterns using a hierarchy of neural
oscillations. This is the concept implemented in
PME. PME represents patterns by cycles (sometimes
called attractors). These cycles occur
naturally with no computation required other than
sampling the pattern. The patterns learned do not
need to be stored, the process never needs to
compare one pattern with another, and there is no
quantization of pattern magnitudes - in stark
contrast to conventional mathematical approaches.
Similar patterns produce similar cycles which
provides the basis for high speed identification of
similar patterns - solving the most common barrier
for developers in a broad range of applications.
Much of the our previous research was to discover and extract the fundamental simplicity of neural oscillations and implement it for practical benefit. The simplicity of this model accounts for the incomparable performance of PME. A type of simplicity consistent with the theme of Stephen Wolfram's book, A New Kind of Science "simple programs with simple rules can exhibit complex behavior".
Company and Technology History:
The PME has evolved over many years including two successful company startups, 6 patents (4 expired), millions of dollars in research/development and thousands of successful installations. The first company was PPTVision, using cameras for high speed visual inspection of manufactured products. The second was AutoData Systems, for recognizing hand printed information from scanned forms. I was the research scientistand founder of these corporations.
These early versions of PME exhibited a small degree of similar pattern clustering, however it was not easily controlled or adjustable. Research in the past 5 years identified parameters to easily adjust the range of similarity without compromising speed and probably represents the most important breakthrough for PME. This discovery initiated the formation of Pattern Memory Inc. along with additional patents pending on the similarity clustering.
Pattern Memory, Inc. offers PME as a tool for developers. Focused on one applications barrier: high speed identifications of similar patterns in noisy raw data. PME basically converts a block of conventional computer memory into pattern memory, (the origin of the company name) allowing developers to use conventional computers to enhance their pattern recognition applications.
Pattern Memory Inc. and PME offers corporations new opportunities to enhance existing or create new products. PME provides the building block for truely intelligent learning machines of the future.
Much of the our previous research was to discover and extract the fundamental simplicity of neural oscillations and implement it for practical benefit. The simplicity of this model accounts for the incomparable performance of PME. A type of simplicity consistent with the theme of Stephen Wolfram's book, A New Kind of Science "simple programs with simple rules can exhibit complex behavior".
Company and Technology History:
The PME has evolved over many years including two successful company startups, 6 patents (4 expired), millions of dollars in research/development and thousands of successful installations. The first company was PPTVision, using cameras for high speed visual inspection of manufactured products. The second was AutoData Systems, for recognizing hand printed information from scanned forms. I was the research scientistand founder of these corporations.
These early versions of PME exhibited a small degree of similar pattern clustering, however it was not easily controlled or adjustable. Research in the past 5 years identified parameters to easily adjust the range of similarity without compromising speed and probably represents the most important breakthrough for PME. This discovery initiated the formation of Pattern Memory Inc. along with additional patents pending on the similarity clustering.
Pattern Memory, Inc. offers PME as a tool for developers. Focused on one applications barrier: high speed identifications of similar patterns in noisy raw data. PME basically converts a block of conventional computer memory into pattern memory, (the origin of the company name) allowing developers to use conventional computers to enhance their pattern recognition applications.
Pattern Memory Inc. and PME offers corporations new opportunities to enhance existing or create new products. PME provides the building block for truely intelligent learning machines of the future.