Suggestion

Hadas Zies (hzies@cs.berkeley.edu)
Mon, 22 Dec 1997 18:44:54 +0100 (MET)

To: BISC Group
From: L. A. Zadeh

The following for your information:

December 10, 1997


To: EECS Faculty
From: L. A. Zadeh
Re: Suggestion of a new 294 series course on Recognition Technology (RT)


Courses on pattern recognition, pattern classification, image
processing, etc. have been around for many years. What is suggested very
briefly in the following is quite different in both form and substance.
After a long period of slow progress, recognition technology has entered
a new phase. Major advances in sensor technology, sensor data processing
technology and recognition methodologies -- especially in neurocomputing,
reinforcement learning and soft computing -- are making it possible to design
and build recognition systems which can do what could not be done just a few
years ago. Here are a few examples.

1. Computer virus detection systems (US Patent 5, 675, 711 assigned to G.
Tesauro et al of IBM). This system employs a neural network classifier which is
trained to detect both known and new viruses.

2. Eyeprint identification in ATM cash machines. In this system, which was
developed by NCR and is being tested in UK, a camera captures a digital record
of a user's iris and can verify identity within seconds from a central database.

3. Checkout scanner which uses scent sensors to identify fruits and vegetables
(US Patent 5, 673, 089).

The point of these examples is that major advances in sensor technology
in combination with comparable advances in machine learning techniques are
transforming recognition technology into a field of major importance on all
levels. What this suggests is that recognition technology should be assigned a
more important place in our curriculum. The suggested course -- which would be
primarily project-oriented -- would be a step in this direction. To teach an RT
course which is on the frontiers of the field would be a real challenge, but it
is a challenge that could and should be met.