resume

Handwriting Recognition Project Overview

First of many stages in
handwriting
recognition

• Fact 1: No computer-based handwriting recognition system has ever achieved human performance, in general.

• Fact 2: Excellent handwriting recognition requires extensive contextual knowledge & understanding.
Example: "the loop in handwritten an e (E) and an e (L) can be identical, graphically. The viewer must understand the context.

• Fact 3: The human neocortex has a fundamentally uniform architecture for speech, vision, touch and motor control,
in which self-organized maps emerge.

• Therefore, Bill and his Motorola team studied the human neocortex, in depth and focused on innovations inspired by the neocortex, such as connectionist speech recognition and handwriting recognition, both of which were demonstrated (with moderate success).

• However, it soon became clear that connectionist (and conventional HMM) approaches lacked hierarchy, flexible representations of time and predictive power. (See PSIQ for Bill's newest solution).

Significance of this Accomplishment

• It avoided millions of dollars being wasted on short-sighted handwriting recognition R&D.

• It consolidated of multiple lines of R&D (handwriting, speech and gesture recognition) into one.

• The neocortex-inspired R&D resulted in a series of highly profitable innovations, such as CorDex.

• Next up is PSIQ.

Skills Transferable to Your Projects:

Leadership / Management
• Bill led the entire connectionist handwriting recognition project.
(Dr. Sidney Garrison, a member of Bill’s team, was an outstanding contributor, too).

Theory / R&D
• Bill focused his team on Neocortex-inspired SOMs, despite industry-wide pressures to focus on HMMs.
• This includes Bill’s numerous insights in machine learning, neuroscience, speech recognition, handwriting recognition, psychology, psychophysics, cognitive science, and computer science.

Software
• Bill led the team’s simulations of numerous connectionist and HMM recognition architectures.

Hardware
• Bill pioneered all-digital and mixed digital/analog hardware designs for demonstration handwriting recognition systems. Cyclone hardware was designed to accelerate such systems. Bill explored Novel memory circuits and materials (e.g. floating gate, FRAM, QRAM ) in order to emulate cortical synapses.

Patents

Patent   (5, 632, 006 Issued   05-20-1997)
Patent   (5, 430, 830     Issued   07-04-1995)
Patent   (5, 216, 751     Issued 06-01-1993)
Patent   (5, 097, 141     Issued   03-17-1992)
Patent   (5, 065, 040     Issued   11-12-1991)

Current Owner: Motorola, Inc.
Licensing availability: (Please contact Motorola).
Contact Info: motorolaventures@motorola.com

Future Opportunities

By pushing beyond the current state-of-the-art,
there are these opportunities:

• The human brain (specifically the neocortex) contains many hundreds of self-organized maps
and there are numerous valuable humans skills (such as speech recognition)
that still defy computer implementation.

• We can learn from nature’s brilliance, or we can ignore it.
Advanced SOM-like hardware learns from it.

• Speech recognition, handwriting recognition and other important human specialties will finally make
great progress, once the "fundamentals" (e.g. neocortical theory) are truly understood.

Links to related material

Patent   5, 632, 006
Patent   5, 430, 830
Patent   5, 216, 751
Patent   5, 097, 141
Patent   5, 065, 040