 |
 |
 |
 |
| |
|||||
|
|
|
Home > News & Events > MITRE Publications > The MITRE Digest > | |||||||||||||||||||
|
Unknown Inside: MITRE's Insights into the Human Brain May 2008
The human brain has peered into the dim fringes of space in order to understand black holes, back into the infancy of time to understand the Big Bang, and into the nanoverse of atomic structure to understand quantum physics. But to gain insight beyond what some describe as "the last frontier of science," the human brain will need to peer into the most mystifying realm of all: itself. Advances in technology that allow researchers to "see" into the brain, by recording and analyzing its electrical and chemical processes, have flooded the field of neuroscience with a bright new light of understanding. MITRE's Neurotechnology Thrust, which is part of our Emerging Technologies Office, aims to focus that light on the problems of MITRE's sponsors. And Brad Minnery, group leader for the company's Neurotechnology Thrust, is excited about the challenge. "Since selecting artificial neural networks as a project topic in an undergraduate computer physics class," says Minnery, "I've been fascinated with the question of how is it that human beings, and even animals, are capable of such remarkable feats—feats that we have not come even close to emulating in artificial systems—using only these fairly sloppy computing elements called 'neurons'?" After completing his Ph.D. in neurobiology, Minnery was happy to find a home at MITRE where he could pursue this question. A Different Breed of Scientist The brain functions differently in many ways from artificial information processing systems. It achieves its amazing processing power in part through the ability of each if its computational elements, called neurons, to form multiple connections with neighboring neurons, and for these local networks in turn to form connections with other networks distributed throughout the brain. While most CPUs process information serially and may fail catastrophically with the loss of even a single transistor, each of the brain's billions of neurons may be connected to up to 10,000 of its neighbors. This massive interconnectivity allows the brain to process information rapidly and in parallel, and it also endows the brain with a high degree of fault tolerance. In the same vein, neurobiologists may differ from the caricature of the "typical" scientist. Minnery believes that the "into the looking-glass" nature of the brain studying itself attracts a particular kind of scientist. "If you circulated through a cocktail party full of scientists, you'd be able to pick out the neuroscientists simply from the poetic bent of their conversation," says Minnery, who himself earned dual degrees in literature and physics. "While other scientists might approach a problem from an applied engineering perspective, neuroscientists tend to view things from a more philosophical vantage." But despite the sometimes esoteric nature of neuroscience, MITRE's research into the subject is completely devoted to the practical need to solve the pressing problems of its sponsors. MITRE's neuroscience activities on behalf of its sponsors can be divided into roughly two categories: bio-inspired artificial intelligence (AI)—including both software- and hardware-based approaches—and human-computer interaction. Artificial Intelligence Back when researchers understood little about how the brain processed and acted upon information, AI programmers attempted to mirror these functions in computers. But million-dollar supercomputers—brutally efficient at number crunching on a googol-sized scale—are still incapable of the adaptive, general-purpose learning already mastered by the brain of a toddler. Minnery and his group are hoping that they can use the burgeoning data on the workings of the brain to help invigorate the field of artificial intelligence. "The story of AI," he says, "is that the grand predictions made for it at its birth in the 1950s have yet to come to fruition. We're several decades down the road, and, despite what we've seen in Hollywood movies over the years, we're not much closer to a building an intelligent agent." Part of the reason for this shortfall of achievement, Minnery explains, is a fundamental misunderstanding of what constitutes intelligence. "While the early researchers in AI concentrated on symbolic reasoning—language, mathematics, playing chess—symbolic reasoning is only the tip of the intelligence iceberg. There are basic cognitive functions, such as perception, attention, learning, and memory, operating beneath the surface of which researchers are only now beginning to gain an understanding. Visual perception in particular is one area where neuroscience-based approaches have been illuminating. It also happens to be a topic of keen interest to our sponsors, many of whom would benefit greatly from the ability to offload mundane image analysis tasks to automated systems." Furthermore, neuroscientists currently have a poor understanding of how the brain's multiple cognitive subsystems interact as a seamless whole. Minnery and his colleagues in MITRE's Neurotechnology Thrust believe that by puzzling out how the separate pieces of human intelligence fit together, we can better replicate it artificially. By emulating neurobiological algorithms and design principles, artificial intelligence programs will be able to adapt better to the changing environmental variables and task constraints with which our sponsors contend in their most challenging projects. Next-Generation Computing Hardware Have you ever wondered how well a three-pound laptop equipped with billions of parallel processors that runs off only 20 watts would sell? Silicon hardware is as yet incapable of matching the capabilities of the brain's "wetware," but there's much computer designers can learn from the brain's information-processing architecture. "There's inherent parallelism in the way the brain computes," says Minnery. "To build a fully functional neural cognitive architecture, you'd need to replicate that parallelism. And parallel processing is just not something that today's serial microprocessors are designed for." A new class of silicon chips, called "neuromorphic chips," attempts to achieve the flexible, low-power consumption performance of the brain by mirroring the connective patterns of neurons as well as their inherent "plasticity"—their ability to modify their connectivity based on changing environmental inputs. But the physical constraints of current hardware design limit neuromorphic chips to approximately one million "neurons" compared with the 100 billion found in the human brain. MITRE's Neurotechnology Thrust, working with Dr. Shamik Das, a computer architect from MITRE's Nanosystems Group, is helping our sponsors move beyond current hardware constraints to develop neuromorphic technologies that incorporate nanoelectronic components. This "nano" construction will allow designers to densely cover a chip with "neurons" and limit the energy that neurons expend communicating with one another. With this approach, it may be possible to build computing hardware that matches the brain's processing agility and energy requirements neuron for neuron, watt for watt. Such hardware will provide our sponsors with an exponential increase in information processing ability and flexibility. Human-Computer Interaction No matter how many synaptic connections its processors possess, a computer's most important connection is to its user. MITRE believes that the ability to model the workings of the brain will lead to breakthroughs in computer interface technology. For instance, one of the Neurotechnology Thrust's research projects, "Biologically Inspired Cognitive Models for Advancing the Design of Command & Control Systems," employs models of the brain's vision and memory systems to design and test display interfaces that are optimized to match the quirks and constraints of human cognition. "There's a class of tools whose purpose is to help commanders visualize what's happening in the battle space—where force elements are in the field, how the situation is unfolding dynamically," Minnery explains. "To help in the design of these tools, we're exploiting neuro-computational models of attention and memory." For instance, Minnery and colleague Julia Hiland are leveraging brain-based models of visual attention to predict the regions within a display that are most likely to draw the viewer's eyes. In order to save time and money, MITRE is attempting to take these and other models, including models of the brain's learning and memory subsystems, and use them to predict how well a given command and control visualization tool will actually succeed in the field. These neuro-computational models may not only help design command and control systems, but test them too. This will reduce the need for the costly and time consuming use of human testing subjects, allowing the effectiveness of new interface systems for our sponsors to be measured more quickly and accurately. New Solutions to Stubborn Problems Peering beyond the borders of neurobiological knowledge will never fail to fascinate the members of MITRE's Neurotechnology Thrust, but it is the applications of MITRE's research in the field that excite them the most. According to Minnery, "Neuroscience is an exciting discipline that's
undergoing radical advancement. There are new insights emerging every
day regarding how the brain processes information. At the same time, there
is a whole class of sponsor problems involving intelligent information
processing that have resisted solution by conventional approaches. By
applying insight from the research, methods, and techniques of MITRE's
neuroscience program, I am confident that we'll make great progress
into solving those problems." —by Christopher Lockheardt Related Information Articles and News
Technical Papers and Presentations
Websites |
||||||||||||||||||||
| Page last updated: May 16, 2008 | Top of page |
Solutions That Make a Difference.® |
|
|
