[comp.newprod] Adaptive Solutions Announces CNAPS System

john@adaptive.asi.com (John Haynes) (03/06/91)

		Adaptive Solutions Announces CNAPS System;
	   	  Speeds Neural Net Learning 1,000 Fold,
	  	   Solves Pattern Recognition Problems

BEAVERTON, Ore -- March 4, 1991--Adaptive Solutions, Inc. today
introduced its CNAPS System, the first system based on the company's
Connected Network of Adaptive ProcessorS (CNAPS) neurocomputing
architecture.  The advanced neurocomputer will speed learning in neural
networks 1,000 fold.  Even compared to a Cray 2 supercomputer, the
CNAPS System executes industry-standard back-propagation algorithms
more than 100 times faster.  The system will provide application
developers with an extremely powerful capability for the solution of
pattern recognition problems in areas such as optical character
recognition (OCR), machine vision, speech recognition, robotic and
process control and financial forecasting.

The CNAPS System consists of a CNAPS server, a neurocomputer for a UNIX
network designed to provide the high speed required for both training
and execution of real-world applications, and CodeNet, a robust
software development environment.

"Adaptive's initial focus is to enable the widespread use of neural
networks in mainstream pattern recognition applications," said John
Heightley, Adaptive president and Chief Executive Officer.  "Our
technology will be used in applications ranging from handwriting
recognition to process control."

"Neural networking will be a key technology in the 1990s," according
the Dan Hammerstrom, the company's founder and Chief Technical
Officer.  "Neural networks will be a pervasive technology.
Applications will expand from high-end industrial systems in
development today to widespread consumer systems in the future.  We
have only begun to scratch the surface."

CNAPS Architecture in Prototype Applications and in Silicon

As an early indication of the potential for Adaptive Solutions' CNAPS
architecture, in November 1990, the company's current industrial
partners, Mitsubishi Electric Corporation and Sharp Corporation
announced independent development efforts in the area of Kanji OCR
based on Adaptive's CNAPS architecture.

Adaptive first exposed details on the chip architecture at the
International Joint Conference on Neural Networks in June 1990.
Details of the first chip based on the CNAPS architecture, the N64000,
were released by Inova Microelectronics, Inc. in February 1991 at the
International Solid-State Circuits Conference (ISSCC).  The N64000 chip
will be incorporated in Adaptive's CNAPS server.

The CNAPS System

The CNAPS System consists of the CNAPS server and the CodeNet software
development environment.  The CNAPS server has 256 Processor Nodes
(PNs) operating in a single instruction multiple data (SIMD) design.
Because each of the processing nodes is a complete digital processor,
complex computations can be performed with massive parallelism.
Broadcast interconnect provides cost-effective, high performance
communication amoung PNs.

The deskside system is a compact unit, measuring 22 inches x 22 inches
x 4 1/2 inches.  The CNAPS server links to a UNIX network and Sun
Sparcstation via a standard Ethernet connection.

Adaptive Solutions expects the CNAPS server to the the world's fastest
learning neurocomputer, capable of exceeding supercomputers such as
Cray Research's Cray 2 and Thinking Machine's CM-2 in raw performance
on neural networking applications.  The CNAPS server will run in
learning mode at more than one billion connection updates per second
(CUPS).  To put this in perspective, Adaptive Solutions' system will be
capable of training NetTalk, a text-to-speech processing network and
the de facto neurocomputing benchmark, in six seconds, compared to more
than four hours on a SPARC workstation.  Applications developers will
be able to get virtually instant turnaround during network training
rather than wait hours for an update.  Peak performance in feedforward
execution mode is 5.12 billion connections per second (CPS) as compared
to the recently announced HNC Balboa 860 coprocessor with peak
performance quoted at 25 million CPS.

CNAPS Software

CodeNet, the CNAPS software development environment, includes the CNAPS
Programming Language (CPL) assembler, the CNtool graphical interface,
and debugger, and a library of common neural network algorithms.

CPL is a modular, extendable, parallel programming language which
allows programmers to build small functional blocks of code and combine
them to make larger programs.  CPL gives the programmer full control
over data and program flow while allowing low-level access to the CNAPS
hardware.

CNtool provides a choice of user interfaces for the CNAPS System:  a
point-and click interface for direct interaction with the system; a
command-line interface for batch processing; and a C library
Application Program Interface (API) for easy programming access in an
embedded application.

The CNAPS System can be programmed to run any learning algorithm.  The
library includes Back Propagation (BP), Learning Vector Quantization
(LVQ), Self Organizing Map (SOM) and Frequency Sensitive Competitive
Learning (FSCL).

Adaptive Solutions will also offer CNAPS-C, a C compiler specifically
designed for the CNAPS architecture.  It has extended constructs to
handle the massive parallelism and scaled arithmetic of the CNAPS
architecture.  The compiler will produce CNAPS executable code and C
code.

Price and Availability

Volume shipments of the CNAPS System are scheduled for the fourth
quarter of 1991.  A limited number of systems will be available to
industrial partners and beta sites in the summer of 1991.  The complete
CNAPS System will carry a U.S. list price of $55,000.  The C compiler
will be priced at $950.  

Adaptive Solutions, Inc.

Founded in 1988 by Dan Hammerstrom, a leader in the field of VLSI
neurocomputing, Adaptive Solutions' goal is to provide neurocomputing
technology for solving problems that cannot be effectively addressed by
traditional computers.  The company has developed a general-purpose
neurocomputing system that has the power, flexibility and adaptability
to make this possible.  The company is funded by Institutional Venture
Partners (Menlo Park, Calif.) and Crosspoint Ventures (Los Altos,
Calif.).

###
For more information contact:
Adaptive Solutions, Inc.
John Haynes
john@asi.com
{uunet,ogicse}!adaptive!john
503/690-1236

Waggener Edstrom				
Alison O'Brien, Susan Pierson
503/245-0905