Brainstorming About Brains at a High-Intensity GatheringMay 2018
Topics: Artificial Intelligence, Machine Learning, Computing Methodologies, Systems Engineering
Who wouldn't want to spend July in the mountains of Telluride talking about how computers can replicate the best of the human brain?
That's where MITRE’s Guido Zarrella will be. As the director of the Institute of Neuromorphic Engineering, he is overseeing the Telluride Neuromorphic Cognition Engineering Workshop. The workshop draws people from around the world to collaborate on the best ways to apply the organizing principles of biologic neural cognition to artificial intelligence (AI).
The Institute is a nonprofit organization that fosters a network of interdisciplinary researchers and scholars working on neuromorphic theories, models, circuits, and systems. It provides tools and support for networking, education, publications, and sharing of resources.
As part of its educational mission, every year the institute holds workshops that bring together people from academia, industry, and government to work on specific problems in intense three-week sessions. This is the 25thanniversary of the Telluride workshop, which is sponsored by the National Science Foundation and other government agencies.
“The work of this group—which involves autonomy, machine learning, and artificial intelligence—grows more important as these technologies progress,” says Zarrella, who is a principal AI engineer with a background in neural networks, language processing, and biological cognition.
Where Humans Influence Machines—and Vice Versa
"The topics of the workshop are critical to our sponsors and their missions," he adds. At a past workshop, for example, he helped create spiking neural networks for IBM's new neuromorphic chip.
"As a result, IBM loaned MITRE a prototype chip to evaluate and test. We wrote papers with IBM on the chip’s potential and eventually developed a work program around these MITRE competencies to help create roadmaps for government applications.”
At this year's workshop, Zarrella is leading a topic with scientists from the University of California, Irvine, and Intel’s Neuromorphic Lab. "We want to design cognitive agents that can learn in the wild. For example, we’ll pioneer new uses for Intel's new neuromorphic chip, Loihi, such as teaching Alexa [Amazon's smart speaker] to learn that it’s being spoken to without a verbal cue."
Past workshop topics have included autonomy and drones, neuromorphic cameras, brainwave decoding, and space situational awareness sensors. "We work on topics that are important to the interaction between humans and machines," he adds.
"For example, last year we looked at compound vision. Tiny fruit fly brains can see predators, perform flight control, and do other complex tasks at the same time. How can we adapt that competency into a computer system?
"The brain can do amazing things, all fueled by a low power source—about 20 watts of energy, the equivalent of one high-efficiency lightbulb. That’s all it takes to run our body functions … what we see, hear, smell, and think. Computers need much more energy to 'think' and act. Can we apply lessons learned from our brains to build more efficient computing systems that are small, light, use little power, but get big results?"
Solving Problems Through Intense Bursts of Activity
MITRE and our sponsors benefit in many ways from our involvement in the institute and the workshops. The benefits include:
- Collaborating with experts from around the world, building new partnerships and networks
- Gaining insight into new technology, ideas, and tools
- Supporting our sponsors’ interest in these topics; sharing knowledge of government needs with researchers
- Bringing back knowledge to MITRE for use on future projects
- Introducing students to the company and its mission.
"I encourage MITRE staff to get involved in the workshops as much as possible," Zarrella says. "There's a great need for this kind of expertise across our work programs."
Overseeing this year's workshop is just one of many items on Zarrella's plate. He already supports numerous sponsors and leads the DEEPLANG research project, which builds enabling technologies that teach machines to fluently understand and generate human language. "We're trying to embed common sense in systems so they can interpret what we mean, not just what we say," he explains.
Zarrella is also interested in recreating the collaborative approach of the workshop at MITRE—that is, pulling a diverse team together for intense bursts of activity to solve a specific problem.
"By bringing together a group of people with different specialties, we produce some exciting collaborations by working on specific problems in short bursts of intense concentration."
—by Beverly Wood