B.U. Bridge is published by the Boston University Office of University Relations.

Contact Us

Staff

Seeing the forest -- and the trees
CNS shares $3.5M grant with Johns Hopkins to model how the brain works

By Brian Fitzgerald

Out for a walk in the rain, whistling your favorite Beatles tune, you freeze when you see a leopard stroll in front of you. It's 40 yards away, but you keep perfectly still because you're not in a zoo, you're in the jungle. The leopard doesn't see you, and that's the way you want it. But what made you stop whistling? How did you see this predator? And, more important, how were you able to separate the moving animal from the movement of the jungle cover in the rain?

Stephen Grossberg. Photo by Kalman Zabarsky

In his office at 677 Beacon St., Stephen Grossberg asks these and similar questions. He has also answered them -- and many others -- through his research on how the brain sees, learns, recognizes, and tracks objects and complex scenes. Now Grossberg, chairman of the CAS department of cognitive and neural systems (CNS), will be better able to establish models on how these brain processes work, thanks to a $3.5 million grant from the Office of Naval Research.

The Multidisciplinary University Research Initiative (MURI) grant will help form a new MURI Center for Intelligent Biomimetic Image Processing and Classification involving colleagues at CNS and Johns Hopkins University, in addition to eight other cooperating laboratories. "The research will further develop a general-purpose, autonomous neural system for vision, object recognition, and tracking applications," says Grossberg, the principal investigator of the grant. "Ultimately, the goal is to apply the research to image processing and pattern recognition applications in technology."

The grant comes a year after Grossberg and other CNS researchers achieved a major breakthrough in modeling how the cerebral cortex processes visual information through its six layers. "This lends itself to technological transfer because it provides a way to deal with the types of uncertainty and unpredictability that are characteristic of a military operational environment," says Grossberg. Vision systems that can more effectively counter an adversary's camouflage, concealment, and deception techniques would be invaluable, especially if they can track multiple targets and have all-weather and day-and-night capabilities. Battlefield and aerospace surveillance would also be greatly improved. Moving ships would be able to better detect moving aircraft.

To be sure, the uses for such research aren't limited to military operations. In the past, brain/behavior models from CNS have been developed by a number of companies, hospitals, and national labs to process data from radar, multispectral infrared light, night vision systems, nuclear magnetic resonance imaging, and high-altitude photography. For example, Boeing's use of adaptive resonance theory (ART), which was developed by Grossberg and CNS and CAS Mathematics Professor Gail Carpenter in 1987, saved the company more than $80 million in airplane design costs. Now, when Boeing wants to design a new plane that calls for thousands of new parts, an engineer can go into a database -- instead of designing a new part from scratch -- and find a part very much like the one needed and, simply by modifying the design, use a part that has already been fabricated.

An ART neural network consists of a short-term memory that captures stimuli, a long-term memory that stores learned information, an attentional subsystem to focus attention on important features of input patterns, and a reorienting system to keep the long-term memory from learning irrelevant patterns.

Grossberg, who was class valedictorian in high school, has been studying neural networks and artificial intelligence for more than 40 years, since he was an undergraduate. "Neural networks are systems that are capable of sophisticated computations similar to those that the human brain routinely performs when neurons transmit signals," he says. Grossberg developed his first neural network during his sophomore year at Dartmouth College, where he again was valedictorian. He received graduate training at Stanford University and earned his Ph.D. at Rockefeller University. After serving on the faculty at MIT, he came to BU and founded the Center for Adaptive Systems in 1981.

In 1988, Grossberg established CNS to provide advanced training and research experience for graduate students interested in the neural and computational principles, mechanisms, and architectures that underlie human and animal behavior, and the application of network architectures to solve outstanding technological problems. Obviously, the military is interested in research at CNS because of its potential applications for machine vision, which enables computers and intelligent sensors to interpret what they see.

"Our team has introduced the most advanced neural model of how the brain uses real motion and long-range apparent motion to track targets whose optical flow trajectories are discontinuous because of multiple occluding objects or drop-out pixels, which are detected by a moving observer," says Grossberg. "We know how the cerebral cortex can simultaneously separate multiple overlapping targets from each other and their backgrounds, complete the representations of partially occluded targets, and use these completed representations to recognize the targets at multiple positions, ranges, and contexts."

Referring to the leopard example, Grossberg points out that battlefield conditions can obscure a target just as jungle cover hides the camouflaged animal. And though the hiker is in motion, he can see the moving leopard despite the surrounding vegetation. It seems like second nature -- it's in your best interest to spot a ferocious feline -- but the process is a complex one. On a speeding ship, how valuable is a vision system that can identify where an enemy helicopter is heading on a foggy day? "In the jungle, you want to know the direction of the leopard before it eats you," Grossberg says with a smile.

16 March 2001
Boston University
Office of University Relations