Personal tools
News from ICTP 101 - Features - Hopfield
Dirac Medal winner, John J. Hopfield, has brought his skills in physics to the world of neurobiology as part of a larger effort to better understand how the brain thinks.
Making Things Compute
John J. Hopfield
Most scientists count themselves
fortunate to acquire international status in just a single field
during their careers. John J. Hopfield, leader of computational
neurobiology and computing networks at Princeton University, USA,
and winner of the ICTP
Dirac Medal in 2001, has had the rare good fortune to distinguish
himself in two fields.
In 1969, just nine years after earning a doctorate in physics
from Cornell University, Hopfield received the Oliver E. Buckley
Prize for his research on the emission and absorption of light
by semiconductors, a topic of central importance for understanding
how light-emitting diodes function. At the time he was honoured,
the prize had been awarded to no less than nine scientists who
had received (or subsequently would receive) the Nobel prize.
At the age of 36, Hopfield seemed well on his way to an illustrious
career in his chosen field.
But over the course of the next two years, Hopfield would switch
his research focus from physics to biology, earning an international
reputation for his pioneering applications of physics-related
computational techniques to the emerging field of neurobiology.
It's the bifocal quality of Hopfield's career--his uncanny ability
to envision and then apply techniques learned in his study of
physics to problems in neurobiology--that led the committee members
of the Dirac Medal to speak of Hopfield's "special and rare
giftto cross inter-disciplinary boundaries to discover new questions
and propose answers that uncover the conceptual structure behind
experimental facts."
"My mother and father were both physicists," explains
Hopfield. "In fact, in the year I was born--1933--my father
was helping to design a physics exhibition at the Century of Progress
World's Fair in Chicago. I can't say I remember the exhibit, but
my father's job at the Chicago world's fair shows that I was not
the first person in my family to use training in physics in unusual
ways." Hopfield made these remarks during a visit to Trieste
this May to receive the Dirac Medal and present the Dirac Lecture.
"As a child in a household filled with physics," Hopfield
notes, "it should come as no surprise to learn that I was
fascinated by the physical world around me. But since the time
that I knew enough to pursue 'new' science, I have always been
more interested in trying to understand things that were not understood
at all rather than in exploring established areas of knowledge
where unanswered questions, by definition, focus on higher level
problems."
Although Hopfield's switch from condensed matter physics to biology
may appear to be a radical new career path to some, for him it
represented a logical extension of the methodologies that he had
been using in condensed matter physics.
"While I was intensely involved in my physics research,"
he says, "I noticed that many biologists were turning to
quantitative measurements common in the study of physical structures.
Put another way, biologists were trying to understand properties
on the basis of structure-an approach that had long characterised
the work of physicists."
Hopfield's breakthrough contribution to neurobiology is based
upon his model of neural processing that offers keen insights
into the vastly different mechanisms that the human brain and
the digital computer use to compute information and make decisions.
The Hopfield model, in fact, demonstrates how qualitatively different
computation in our brain and in a computer can be.
"The secret behind these differences," Hopfield notes,
"lies in connectivity. In the human brain, each neuron makes
'synaptic' connections to thousands of other neurons in a vast
and intricate network. In a computer, despite its complexity,
each transistor is usually connected to only two or three other
transistors. The human/computer 'connectivity divide' represents
not just a quantitative difference but a fundamentally qualitative
difference that affects the way in which decisions are made."
Hopfield is particularly interested in the way in which the human
brain understands the world through its senses. "Every sensory
system," he observes, "senses--or, put another way,
makes sense of--the world by dividing it into objects. Our visual
system does it, our auditory system does it, and our olfactory
system does it. Working in tandem with our sensory systems, the
brain's ability to separate and objectify the world is what makes
the world coherent."
"All of our sensory brain interactions involve complex physical
structures and correlations," Hopfield notes, "but the
visual system is particularly specialised and complicated."
Except for deciphering colours, the retina functions through a
large number of identical receptor cells. "When the retina
is exposed to light," he continues, "the pattern of
excitation of the retinal cells is configured only when a given
object is present (for example, a face). The pattern, however,
also depends on the exact direction in which we are looking."
Consequently, the activity of a retinal cell is not only determined
by what we are looking at but how we are looking
at it--a situation that presents enormous complications for the
study of visual pattern recognition.
By contrast, our sense of smell is based on having about 1000
different types of receptor cells, each harbouring a different
response to any given odorant. As a result, in most circumstances,
the most strongly driven receptor cells are determined solely
by the object being smelled.
"Olfaction, "Hopfield observes," is one of the
oldest and simplest senses. That has made it a logical place to
pursue my neurobiology research and it explains why so much of
my work has been based on studies of the olfactory system."
"The research challenge," he adds, "is particularly
interesting in animals that use olfaction for remote sensing--for
example, such carnivores as hyenas and bears that apply their
sense of smell not only to identify and locate objects right in
front of them but also objects that are not near-at-hand. Such
remote sensing involves an assessment of the direction and force
of the wind as well as an ability to distinguish the odour of
the 'targeted' object from the background odours that are also
present."
It's the nasal equivalent of being able to identify a distant
sound by filtering out all of the noise in between. The ability
of pigeons to find their way home over vast distances and the
capacity of slugs to pinpoint their favourite food while sliming
along in your garden provide excellent examples of such remote
sensing capacities.
Hopfield recently discovered another organising factor in olfaction
and, in the process, demonstrated a new principle explaining how
the neural function takes advantage of the 'spiking' phenomenon
(characteristic of interneural communication) in carrying out
its computations. "My current research," he says, "focusses
on how the brain's neural circuits produce such powerful and complex
computations."
Hopfield acknowledges that understanding the biophysics of neurons
is an enormously complicated task. How can we visualise--let alone
begin to understand--the intricate web of biological and physical
factors that is responsible for the way in which sensory information
is presented to and then deciphered by the brain?
"Light and chemo-reception," he notes, "generate
currents across a cell membrane by means of a cascade of physical
and chemical events within a receptor cell. When information leaves
the eye, nose or ear, it is represented as a sequence of action-potentials.
Greater understanding of what turns action-potentials into objectified
realities would shed enormous light on how the brain transforms
sensory inputs into coherent information."
Hopfield believes that the 21st century will be the century of
neurobiology much like the middle half of the 20th century was
the century of the atom. "Researchers have been developing
increasingly sophisticated techniques for understanding the basic
mechanisms by which the brain computes, a process that has largely
involved understanding the brain in computational terms."
"One of the most intriguing aspects of recent research,"
Hopfield notes, "has been the interaction--the intellectual
back and forth--between trying to comprehend the workings of the
brain and trying to develop brain-like computer programmes for
useful application."
"As I mentioned before, the brain's intricate circuitry,
driven by neurons linked together with huge synaptic connectivity,
means that a brain functions differently than a computer. Nevertheless
much of what the brain does can be described in terms of computation:
associative memory, logic, inference, generating an appropriate
sequence of locomotive commands, recognising a distinctive odour
or judging the precise location of an object. All of these functions,
at a fundamental level, require computation."
And that's where the application of computational tools, previously
associated with physics to address unanswered questions in neurobiology,
holds great promise. Such cross-disciplinary investigations, which
serve as the centrepiece of Hopfield's current research, may offer
an ideal strategy for better understanding how the human brain
thinks.