CorDex Project Overview

View of clusters and sub-layers in CorDex

• Bill invented a powerful Data Mining Software Suite (called CorDex*)   and led the development of a highly successful commercial product. • This software suite is capable of self-organizing diverse data from large databases.   Once self-organized, the results are presented to the human user as colored clusters.   With a few mouse clicks, the user can drill down through sub-maps to find root causes. • A rule induction feature helps the user find and interpret the results. • Keywords: data mining, self-orgnaizing maps, clustering, rule induction, labeled data.          * The name “CorDex” is a contraction of “Cortex-like-index”  .

Significance of this Accomplishment

• First, Bill’s R&D Group used CorDex to discover root-causes of expensive yield variations in Motorola’s own semiconductor business, thereby saving Motorola millions of dollars. (These causes were extremely subtile, and had baffled a sequence of three experienced product engineers, prior to trying CorDex). • Later, a startup “Global Knowledge Services, Inc.” (GKS) was licensed by Motorola to use CorDex in the same narrow field-of-application (semiconductor yield-cause discovery). About three years later, GKS was acquired by Applied Materials, Inc. for about $40 million. CorDex was the primary tool GKS used for their success. • CorDex is applicable in almost any field-of-application, not just semiconductor yield-cause discovery.  (Note: CorDex is not hierarchical, which would be even more powerful).

Skills Transferable to Your Projects:

Leadership / Management • Bill invented the CorDex concept and had the vision to see its potential, months before any others did. • Bill kept presenting the facts and logical projections until others finally recognized its high value. • Bill led and managed a Talented, Cohesive Team:         Robert Leivian expertly coded CorDex, adding many key usability improvements.         Mike Gardner became our most expert CorDex user, and an excellent leader, too.         Dr. Kari Torkkola added rule induction and accelerated the map software.           John Summers supported all aspects of the project, especially the graphics.     Theory / R&D • As the name implies, CorDex was inspired from the mammalian neocortex, and self-organizing   maps, which Bill has studied for years. • CorDex R&D demonstrated Bill’s numerous insights in machine learning, neuroscience,    computer science.  Software • Bill helped plan the CorDex software and Robert Leivian coded it in C & C++. • (Robert worked in Bill’s group). Hardware • The only hardware aspect of CorDex, aside from the “semiconductor yield aspect”, was the occasional use of Bill’s group’s “Cyclone” (an accelerator for self-organizing maps) for especially large data bases.

Patents

Patent   (5, 919, 245) Issued   (10 - 6 -1998) 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: • As noted above, CorDex is not hierarchical, which would be even more powerful.   See here for an example of hierarchical self-organizing maps. • Even the discipline of “attributing” (defining a good set of numerical attributes) an complex issue has great value, and can become the basis of new powerful neuroscience-inspired tools and products.  Bill has the insight and experience to make it happen.

Links to related material

• Introduction to self-organizing maps   http://en.wikipedia.org/wiki/Self-organizing_map • CorDex WSOM Conference Paper. • Applied Materials, Inc. acquires GKS   http://findarticles.com/p/articles/mi_m0EIN/is_2001_Dec_4/ai_80526396