New this year at GEOINT Foreword, the pre-conference science and technology day that precedes the GEOINT 2016 Symposium, were presentations from four U.S. National Laboratories.

Tracking Illness in the Information Age

Turning to Google to determine what’s ailing you has become a common ritual in the information age. It turns out “Dr. Google,” as some physicians refer to the trend, could in the future help authorities track and control the spread of disease.

Scientists at Los Alamos National Laboratory (LANL) are studying how open-source data such as social media can be used to recognize health-related trends. When people interact through the internet, they often leave clues about their health, said Nicholas Generous, a digital epidemiologist at LANL. That data trail can be combined with traditional disease-tracking methods—including reporting by individual physicians—to better understand health trends.

An uptick in Google searches for flu symptoms is one indicator that can be picked up via big data analytics, Generous said. Another is Facebook and Twitter posts where people mention something health-related. These data points often include location information as well as a time stamp, enabling analysts to place them in a geospatial and temporal context, Generous explained. When analyzed altogether, such information can reveal a lot about the severity and scope of a pubic health concern.

Smartphones and wearable devices such as Fitbits also have great health-tracking potential. These devices carry sensors such as GPS receivers and accelerometers that can precisely track the movements of those carrying or wearing them. This data can be useful in monitoring the progress of degenerative disorders such Parkinson’s disease or determining whether Alzheimer’s patients can still care for themselves, Generous said.

Mental health also has the potential to be tracked via people’s online interactions, Generous said. When someone becomes depressed, for example, their social media usage tends to decline. But health-trend monitoring via the internet is not an exact science, Generous cautioned. For example, the Ebola outbreak of 2014 triggered a huge spike in U.S. searches about the disease, even though the number of confirmed cases in the country was miniscule.

Turning Advanced Radar into Real-Time Decisions

The Intelligence Community is drowning in data from airborne sensors. Fortunately, Sandia National Laboratories (SNL) is building a life preserver, according to Bert Tise, distinguished member of the SNL technical staff.

“We’ve got deployed systems out there that are producing tens of millions of images per year,” said Tise, whose presentation focused on processing, exploitation, and dissemination of broad-area, high-resolution imagery from synthetic aperture radar (SAR). “When you get a data pile that size, what do you need to do to [analyze] it?”

That question keeps SNL engineers up at night, Tise said, but SNL’s answer is a mathematical data analytics approach that turns abstract pixels from SAR imagery into geo-temporal “semantic graphs” that are searchable and node-based.

“These graphs represent the… relationships between the features that are on a map,” continued Tise. The relationships shown on the graphs help analysts reduce the size of their data pool in pursuit of better, faster answers to pressing intelligence questions. “It helps with the needle-in-a-haystack problem,” Tise said.

Tise shared an example from the Federal Aviation Administration (FAA): a national map of all the airplane flights documented by the FAA on a single day in 2014. SNL developed an algorithm based on the geometric features from the map that allowed analysts to identify “odd flights”—outliers that didn’t conform to typical flight patterns.

“With our system … we can narrow down all that data, based on the definition of ‘odd flights,’ to about 700 out of the total 50,000 flights across the U.S,” Tise said. “It lets an analyst get in there, ask very nuanced questions, and quickly spot things that don’t belong.”

The result: turning high-quality images into high-quality decisions.

Mapping Human Population Trends

Imagine being able to visualize every hut in Nigeria, every mobile home in Texas, and all the school children in Philadelphia during recess.

Oak Ridge National Laboratory (ORNL) is able to do this and much more as it ramps up research in three platforms: LandScan Global, LandScan USA, and LandScan HD. All are designed to study the world’s populations with a fine-tooth comb.

Eddie Bright, team leader for the lab’s Population Distribution and Dynamics Team, said ORNL has the ability to stratify urban areas by income levels, types of economic activity, and even occupancy within specific buildings. They can also get information about parts of the country where Census data is scant or unavailable.

“No census worker has ever gone to these areas,” said Robert Stewart, Geographic Data Sciences team leader.

The Lab’s capabilities have been used to gather information on Iraqi refugee movements and assess future population dynamics with respect to trends in climate change and natural disasters.

“If you dump a bucket of people on a terrain, where would they flow?” Budhendra Bhaduri, a corporate research fellow with ORNL, posited. “As a people, we are all programmed to mitigate risk and look for short cuts. We can understand where people would try to migrate during emergencies.”

Using large data sets, from the World Bank, for example, the team can glean 30 attributes of the whole world, and from these, “generate the most iconic trend lines,” Stewart said.

The lab is also researching population via social media. Its PlanetSense program explores the relationship between Twitter and Facebook activity and population trends.

For example, “Does more Facebook usage mean the population is actually rising?” Stewart pondered.

Future Challenges of Big Data and Deep Learning

It’s a known fact for anyone in the artificial intelligence world: We are living in an era of exponentially growing data sets and how we mine them is in part a matter of how savvy and how quick our computers are at pattern recognition.

So far, computers have been the mainstay of extracting information from large amounts of data and providing something meaningful to a human analyst.

“The future is going to be a lot more challenging,” said Paul Metzger, of MIT’s Lincoln Laboratory. “With more and more data, a lot of systems are talking to other systems. Manned platforms are interfaced with unmanned platforms. We need to get to a point to have a variety of systems acting as one.”

Metzger, group leader of Intelligence & Decision Technologies at Lincoln Lab, gave an overview of the lab’s beginnings in 1951, when it was already at the forefront of developing autonomous systems to augment humans, up to its present-day research areas.

“We are working more in terms of disaster response,” Metzger said.

This could mean situational awareness about natural disasters such as California wildfires.

The lab is also focusing on cybersecurity using a system called Cloudbreak.

“How do you give a cyber commander an operational picture of what it looks like in a cyber environment?” Metzger said.

The Lab is also training computers to perform first level data triage.

“Every image [is being viewed by] one set of eyeballs,” Metzger said, adding computers could vet images for abnormalities. “What happens when the computer sees something it hasn’t seen previously?”

For example, when a computer “sees” a new aircraft or ship, he elaborated. Having a triage system in place would allow the computer to detect a discrepancy between the number of ships on a given day versus the norm.

Co-written by Matt Alderton, Kristine Crane, and Warren Ferster 

Related


Posted by Matt Alderton

Leave a reply

Your email address will not be published.