UC Berkeley Press Release
Jeff Hawkins, computing pioneer, endows new center to develop model of brain
BERKELEY – Jeff Hawkins, creator of the first commercially successful handheld computer and author of the book "On Intelligence," has endowed a new research center at the University of California, Berkeley, to develop mathematical and computational models of how the brain works.
The founding of the new Redwood Center for Theoretical Neuroscience, which will operate within the Helen Wills Neuroscience Institute, is being celebrated this Friday, Oct. 7, with a day-long symposium at the campus's Faculty Club.
Hawkins attended the graduate program in biophysics at UC Berkeley nearly 20 years ago to pursue his interest in neuroscience. After his stay at UC Berkeley, he co-founded Palm Computing and Handspring, introducing the very successful PalmPilot and Treo product lines.
Three years ago, Hawkins again turned his attention to neuroscience with the founding of the Redwood Neuroscience Institute (RNI) in Menlo Park. During this time, Hawkins published the well-received "On Intelligence" with co-author Sandra Blakeslee.
With a $4 million gift from Hawkins and his wife, Janet Strauss, UC Berkeley is creating the new Redwood Center. Bruno Olshausen, formerly of RNI, will become an associate professor of neuroscience at UC Berkeley and head of the new center.
The transition of RNI from a standalone research institute to a UC Berkeley entity coincided with the founding of Hawkins' new company, Numenta, Inc., which is developing a new type of computer memory system modeled after the human neocortex.
"The most logical home for RNI is Berkeley, because I had a very close connection with Wills institute director Bob Knight, who had helped me form RNI, who had served on the board for three years, and has been a great partner for us," said Hawkins. "We were thrilled that the Redwood Center could be created under the leadership of Bruno Olshausen of RNI."
"There is a natural affinity between the two groups," Olshausen said. Despite its move under the aegis of the Wills institute, Olshausen and his team will pursue research goals similar to those at RNI: to create a model of the brain that can be tested in the lab and eventually mimicked by a computer.
"Our goal is to develop a theoretical framework for the neocortex - the outer layer of the brain involved in conscious perception and action," said Olshausen. "There is no good framework now, despite tons of data. The field is data rich, but theory poor."
"It's a perfect match for the campus," added Knight, UC Berkeley professor of psychology and of neuroscience. "The center is really going to interact with a lot of people and stimulate a lot of excitement. And Bruno is a superb computational and theoretical neuroscientist, one of the best of his generation."
Knight praised Olshausen and Hawkins as "guys who don't just come up with unrestrained theory. They build theory based on known observations, anatomically or physiologically." Olshausen, for example, originated a theory called "sparse coding" in the late 1990s, proposing that the brain operates with the least number of neurons possible.
"Sparse coding means that if you can reliably transfer information or communicate between brain areas or solve problems with 10 cells as opposed to 100 cells, the brain will try to do it with a limited number of cells, with enough fidelity to have quality information. Use the minimum amount of cells to get the maximum pop," Knight said. "Bruno proposed that in the late 90s and now it's been shown to be true in many systems by other investigators."
Olshausen currently is interested in how the visual area of the brain encodes images that allow us to recall and recognize them in an instant despite different orientations and different lighting.
"If you look at and recognize a pencil, then re-orient it, it's still a pencil," he said. "From a computational point of view, thinking of a pencil as a bunch of pixels, it's an immense challenge to understand - it's still a big problem in computer vision. Yet somehow, in humans, it's done naturally. How does this happen?"
The much-touted area of artificial intelligence tried to solve problems such as this, too, but without reference to the way the brain does it. The area of cognitive science tries to bring psychology to bear on the question, creating mathematical models based on behavior.
Olshausen and his colleagues, however, want to understand the cortex at a level closer to the individual nerve cell so as to explain things such as memory, sensory motor function and attention.
"We're trying to drive down to the neurobiological subunit, including the individual nerve cell," he said. "If we are successful, we should be able to produce a theory that it testable by neurobiologists recording at the level of individual neurons, thereby relating neural activity to perceptual function."
The theories also can be tested on live volunteers using the institute's functional magnetic resonance imager (fMRI), part of the Wheeler Brain Imaging Center in the Wills Neuroscience Institute. Other areas of interest among the six researchers in the center are how the sensory world is encoded in the brain, how memories are stored and how the brain synchronizes activity.
As far as Hawkins is concerned, understanding of the neocortex is already at a level where it's possible to create software that is inspired by the algorithms of the brain, and therefore train computers to do things we would label as intelligent. He is putting his beliefs to action with the founding of his new Silicon Valley start-up, Numenta.
"When it comes to looking at pictures, and saying, 'What is in this image?' 'What am I looking at?' - there is no computer that can do that today," he said. "There is no computer that can listen to speech or read text and understand what is being spoken about. Numenta's technology can do that. There are no good algorithms for general purpose robotics, how to make a machine that can walk around in a fluid way, not like some lumbering zombie. Insects do it, mice do it, humans do it - we don't know how to do that. Numenta's technology can do that."
Like the earliest digital computers, however, the machine Hawkins hopes to build will have applications no one yet dreams of, he said.
"More importantly, the technology can do a lot of things humans don't do well," said Hawkins. "It can understand complex worlds and make predictions about them, and that is the core of what intelligence is all about - forming an understanding about a world and then predicting the future. We have a fundamental technology, which is basically brains, intelligence and the neocortex, but the most interesting applications will be ones that are hard to think of now."
He stressed, however, that the new Redwood Center has lots of work to do fleshing out the theories he's taking to the computer design board.
"There is a lot of work still to be done, mapping the theory to the biology and understanding exactly how the biology produces these things," he said. "There are a lot of details in the biology that still need to be worked on. And that is going to continue. Yet, I have enough now to start a business that builds the technology side of it. The Redwood Center is going to continue doing the neuroscience aspect of cortical theory."
Aside from cortical theory, however, Knight emphasized that the center will help researchers across the campus by providing novel computational and theoretical perspectives to better understand the massive data sets generated today by studies of the brain. Handling monstrous data sets and extracting information from them requires sophistication in both computation and brain theory.
"There are so many ways to look at complex physiological data, you really have to have people who are mathematically sophisticated, and you have to have people who have thought about the theory and how the brain might actually use different patterns of neural firing or other brain patterns to produce behavior," he said. "It's really not the computer so much, it's more how do you understand the problem and extract the signal from the noise and verify that it's real. That's the area where it takes computationally savvy and theoretically savvy people to make a difference."
Hawkins' endowment is designed to provide graduate student support and fund seminars and conferences, such as the day-long inaugural symposium on Oct. 7 that will feature neuroscientists from around the world. Hawkins, too, will speak on "Prospects and Problems of Cortical Theory."