hmm ... anything to do with starlabs?
---------- Forwarded message ----------
Date: Sat, 24 Jun 2000 12:00:11 PDT
From: "UPI / LIDIA WASOWICZ, UPI Science Writer" <C-upi@clari.net>
Newsgroups: clari.tw.science, clari.tw.features, clari.tw.misc,
clari.tw.science+space
Subject: Building brain-like circuits
By LIDIA WASOWICZ, UPI Science Writer
SAN FRANCISCO, June 23 (UPI) -- Inspired by the brain's inner
workings, researchers have built an electronic circuit with a
biological twist.
The scientists from the Massachusetts Institute of Technology in
Cambridge, Mass., Bell Laboratories of Lucent Technologies in Murray
Hill, N.J., and the Institute of Neuroinformatics in Zurich,
Switzerland, report in the British journal Nature their advance in
blending neurobiology and computer design.
MIT's Richard Hahnloser of the Department of Brain and
Cognitive Sciences and Rahul Sarpeshkar, assistant professor of
electrical engineering and computer science, and H. Sebastian Seung,
assistant professor of computational neuroscience, who are also Bell
Labs consultants, and Rodney Douglas and the late Misha Mahowald of the
Swiss center devised an electronic circuit that mimics the biological
circuitry of the cerebral cortex, the brain's center of intelligence.
The team developed a mathematical theory of how analog
amplification and digital selection can co-exist in the same circuit,
a combination once thought unlikely.
"Apparently the brain, and computers wishing to emulate it, may
have it both ways," the study authors said.
The work "demonstrates that we know enough today to begin building
integrated circuits that compute like biology," Chris Diorio of the
University of Washington in Seattle and Rajesh Rao of the Salk
Institute for Biological Studies in La Jolla, Calif., said in an
accompanying News and Views article.
The feat in the emerging field of neuromorphic engineering --
creating devices that have similar neural systems -- could lead to
machines that can recognize objects by sight and perform other
sophisticated perceptual tasks, scientists said.
"Like electronic circuits, the neural circuits of the cortex
contain many feedback loops," Seung said. "But neuroscientists have
found cortical feedback seems to operate in a way unfamiliar to
today's electronic designers. We set out to mimic this novel mode of
operation in an unconventional electronic circuit."
The researchers have a multi-dimensional vision of potential
applications -- though none so bold as the view of the future espoused
by techno-maverick Ray Kurtzweil, who in his latest book, "The Age of
the Spiritual Machine: When Computers Exceed Human Intelligence,"
asserts the machines of tomorrow will be smarter than the people.
"The main use of devices such as ours is for autonomous agents,
such as robots," Hahnloser told United Press International. "Like
humans, they need circuits for specific purposes such as obstacle
avoidance. The fact that the circuits we built have a very low power
consumption compared to purely digital circuits enhances the potential
autonomy of these agents by orders of magnitude."
The silicon technology that has been driving the computer
revolution continues to improve at rates predicted by the so-called
Moore's law, which forecasts a doubling in computer capacity every 18
months, the researchers said.
"But there are many scientists working on novel technology such as
quantum computers, DNA computers and other substrates for computation.
Which one will make it I don't know," Hahnloser said.
"The fact that we have expressed our current understanding of
recurrent cortical circuits in terms of transistors and silicon might
turn out to be important. Transistors have such a long history of
success, I think they will survive in one form or another. Whether it
will be semiconductors or not remains to be seen."
A spate of recent breakthroughs has imbued scientists with a sense
of excitement about witnessing what they see as the start of an era in
digital electronics -- one marked by circuits only a few atoms across
but with unfathomable speed and memory or machines no larger than a
blood cell yet fully programmable and perhaps even able to duplicate
themselves.
Groups at Hewlett-Packard, the University of California, Los
Angeles, and elsewhere working on this ultramicroscopic scale, called
molecular electronics, have already created rudimentary electronic
"logic gates" and other basic components of computing with the
thickness of a single molecule.
The ultimate aim is to create chemical reactions that assemble
gargantuan numbers of circuits at a cost as minute as their molecular
scale.
Other physicists investigating the brain's feats have developed a
program that can recognize patterns, a so-called learning algorithm
that may lead to significant improvements in a host of fields, from Web
searches to voice recognition and other computer applications to
medical imaging.
Brain-inspired circuit technology could have a vast impact,
investigators said.
"A PC is a universal machine, capable of doing about anything you
program it to -- if you have time enough to wait for the answer,"
Hahnloser said, noting the new advance could make a pronounced
difference in how -- and how fast -- such computations are performed.
"If you are about to hit an obstacle, you'd rather recognize it
quickly and trigger an avoidance behavior early," he said.
The brain's basic computing element, the neuron, is on the slow
side and incapable of performing such difficult computations as
recognizing a familiar face from different angles and under different
lighting conditions, scientists said.
Thanks to their dense interconnection, however, neurons can share,
exchange and process information. "Therefore, as a whole, they can
achieve incredible speed, reliability, fault tolerance and capability
of adaptation and learning," Hahnloser said.
"Digital computers such as the PC are different. The memory and the
processor elements are physically separate from each other. Each basic
computation is executed very rapidly, but the computation as a whole is
executed serially, in a step-by-step manner."
And digital computers have no tolerance for error -- delete a
single wire, and the computer may stop functioning altogether.
"In this sense our circuit is more like the brain," Hahnloser told
UPI. "The computation is done in parallel, thanks to the dense
interconnectivity between the many artificial neurons."
While they have not yet demonstrated it, theoretically, the circuit
should still perform fully even with a severed wire, the researchers
said.
"Our work is situated at the direct interface between neuroscience
and engineering," Douglas said. "We use insights from neuroscience to
improve our understanding and creativity in engineering and vice
versa."
The approach in taking inspiration from biology to build circuits
is rooted in the pioneering work of Carver Mead, a professor at the
California Institute of Technology in Pasadena, Calif. All of the
previous work focused on the periphery, rather than on the center, of
the nervous system.
A decade ago, for example, Mahowald built a silicon retina,
implementing the first stages of the visual system. Sarpeshkar later
constructed a silicon cochlea, representing the first stages of the
auditory system. And Douglas and Mahowald devised an electronic
equivalent of a real neuron, emulating its many ionic currents.
"Our goal is aimed at implementing a higher brain function -- not
directly relating to the periphery of the brain circuits -- which is
the computations that are performed in the recurrent circuits of the
neocortex," Hahnloser said.
The new circuit is composed of man-made neurons that communicate
with each other via artificial connections called synapses, all of the
elements being made from transistors fabricated on a silicon integrated
circuit.
The human brain is a vast and intricate network, comprised of
billions of neurons, each of which might be connected to 10,000 or so
others, scientists said.
"Biologists like to focus on simple linear pathways through this
network, ignoring the tangled web of feedback loops, which seem too
complex to even contemplate," Seung said. "But it seems unlikely that
we could ever understand intelligence or consciousness without
understanding the role of feedback in the neural networks of the
brain."
Scientists have tended to draw analogies between electronic and
neural circuits, but recent studies suggest the brain in fact does not
rely on feedback in the same way as do conventional electronics, which
are distinctly analog or digital.
"Analog is the value of a perception, such as the intensity, the
contrast or the color," Hahnloser told UPI. "Digital is the constraints
by which we perceive. We can't look at a face without recognizing it as
such."
Thus, perception combines digital and analog aspects, the study
authors pointed out. When a human sees an approaching car, he also
receives a continuous stream of information about its color, changing
size in relation to its distance, spatial relations to other objects
and the like.
"Philosophers and psychologists have long been struck by the
duality between analog and digital in perception," Seung said. "They
have further speculated about whether the computational operations
underlying perception in the brain are analog or digital."
Their research, the authors concluded, indicates the two sides are
not mutually exclusive; rather they appear to coexist in the brain's
neural circuitry.
The research was funded by the Swiss National Science Foundation
SPP Program, Lucent Technologies and MIT.
--
Copyright 2000 by United Press International.
All rights reserved.
--
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:14:15 MDT