Hometown U: Artificial intelligence in the Arctic
Three undergraduates and an expert in artificial intelligence huddle in a small darkened lab at the University of Alaska Anchorage. On a square table top before them, lights flicker on and off, on and off, in a simple pattern — 1-2-3-4, 1-2-3-4, always heading toward the table’s edge.
The winking lights are attached to $10 off-the-shelf computer chips wired into the pegboard. About 50 little chips are engaged; the blinking path is not always linear. In fact, the lights manage to travel around a chip that is broken or out of service. The expert calls the system “self-healing” because the chips find a path to the board’s edge despite obstacles.
Each chip by itself is a complete computer. Standing alone, it’s pretty dumb. But programmed to talk to each other, this array of blinking lights becomes an intelligent information-gathering and transmission system. Low-powered and capable of functioning in primitive conditions, the network must work easily in environments with no technical infrastructure. Say, the Bering Sea ice pack.
Martin Cenek, a professor and computer science researcher at UAA since 2012, earned his doctorate in artificial intelligence from Portland State University. His fascination with topics like complex systems and machine learning inspires his classroom discussions and continues to influence his research.
The work he’s demonstrating today is funded by the Department of Homeland Security. About this time last year, the DHS named UAA as a Maritime Research Center of Excellence for the Arctic Domain. The center provides funding to develop tools and systems that can “observe, assess, predict and alert stakeholders” to developments in the Arctic. Think oil spills, ocean acidification, port security, ice breakup, fisheries management or vessel traffic.
Over the next five years, $17 million from the federal government will flow through the center for projects that deliver new technologies to advance DHS’s mission, according to its director, Helena Wisniewski.
As ADAC’s home base, UAA partners with 16 other entities — four universities, Woods Hole Oceanographic Institute in Massachusetts and nine industries that run the gamut from Lockheed Martin and Liquid Robotics to the Port of Anchorage. Rural residents in Gambell are employed as community observers, noticing anything from ice movement to unusual ships. The U.S. Coast Guard is the main client for the center’s work.
So what’s all this got to do with Cenek and his blinking lights?
He and his team are creating the low-power, decentralized, asynchronous (each operation starts only after the preceding one is complete) system that can be outfitted with any number of sensors and deployed remotely. Although they’ll power it with solar or wave action, the team figures their system could run six months on four AA batteries.
So what’s special about the chips? They count. Think of a bunch of people standing next to each other. I start off as 0. I tell you 0; you turn around and find the next person and you say 1. That person says 2 to the next person. The message is being changed to keep the count.
“So this will blow your mind,” Cenek said. “Do not think of these as sensors. Think of them as neurons in your brain; they work in a funny way.”
He explained by asking us to be neurons. Cenek, as neuron, spots a bird; he passes the information on to another neuron, say Matt Devins, one of his undergraduate research assistants. If the message arrives while Devins is paying attention, he gets excited and fires off a message to Mike Mobley, another assistant. But if the information arrives outside Devins’ attention span, Devins won’t pass the information on to Mobley.
“This is one model of how the brain works,” Cenek said. “It is called time-dependent neural spiking network. This is what I want to use to drive the intelligence of this network.”
By that, he means he wants the system to be able to differentiate between “ham or spam,” signal or noise. Sometimes, moving ice is just a seal.
But that’s not easy. “There’s a reason no one is doing this,” Cenek said, and both research assistants jump in to explain.
“(The network of chips) are like all these people screaming in an auditorium, all in the same voice,” Devins said.
“At the same pitch and tone,” chimed in Mobley. “They kind of argue with one another and we get undesirable behavior, we miss data. So that’s a been a big challenge.”
Which is why Cenek has been borrowing paradigms from biology, from neurophysiology, “from whatever we can get our hands on just to make this intelligence work in a similar fashion as our brain works.”
He made another point: This work is science, not engineering.
“As engineers, we’d just rig up some hack to get the job done. I would just spend more money and make some of these [chips] more powerful. But that would be skirting our problem. We have to do the low-level science because this has to function in an environment with no traditional infrastructure and supports.”
Kathleen McCoy works for the University of Alaska Anchorage, where she writes about campus life for social and online media.
Via: Google Alerts for AI