10 Bits: the Data News Hotlist
This week’s list of data news highlights covers February 20-26, 2016 and includes articles about a machine learning system that can stop digital bank robberies and a new system that can keep drones from getting too close to airports.
President Obama has announced a series of new projects to accelerate the Precision Medicine Initiative (PMI). Vanderbilt University and Alphabet-owned life science company Verily will study how to recruit and collected data from the million volunteers it plans to enlist in its research, and the Health Resources and Services Administration will develop strategies for engaging underserved and underrepresented populations. In addition, several companies have agreed to support the PMI, such as by developing standardized application programming interfaces to facilitate patient data exchange.
A project developed by the Smart Earth Network (SEN), a technology and conservation organization, and conservation nonprofit C3 is using the Internet of Things to help fishermen in the Phillippines track and protect dugongs, which are believed to be close to extinction due to illegal fishing and habitat destruction. Fishermen equipped with smartphones photograph any dugongs they spot at sea and upload these geotagged photos with an app developed by SEN to a cloud platform so researchers and conservationists can monitor dugong populations and use this data to guide conservation efforts.
MasterCard has credited its new machine learning technology Safety Net with successfully detecting and stopping three separate cyberattacks designed to steal from automated teller machines (ATMs) during the first two months of 2016. Safety Net, which MasterCard launched internationally in late 2015, allows Mastercard to analyze over 1.3 billion transactions per day and automatically identify signs of malicious activity. Safety Net’s algorithms can mine transaction data and detect potential red flags, such as unusually large transactions, and warn banks that they may want to investigate or stop a transaction. Safety Net limited the damage of these attacks to a total of approximately $100,000 each. For comparison, a similar attack in 2013 quickly spread to 26 countries and stole $40 million in 11 hours.
Google has developed a deep-learning system called PlaNet capable of analyzing photos and identifying where in the world it was taken better than humans can. Google trained PlaNet on 91 million images that included geotagged metadata so its algorithms could learn to associate particular traits, such as architecture style or vegetation, which particular locations. Google tested PlaNet by comparing its ability to identify the location of new images to the ability of a team of 10 “well-traveled” humans, and PlaNet won 28 of 50 rounds, successfully identifying the location of images with a median error of approximately 703 miles, whereas the human median error was 1,442 miles.
A predictive analytics program at the University of Iowa Hospitals and Clinics has dramatically reduced the amount of surgical site infections (SSIs) by modeling infection risks with historical and live surgical data. SSIs are among the most common and costly complications that can arise in a hospital and can pose substantial harm to patients. When the hospitals introduced the system in 2012, the system was able to reduce SSIs by 58 percent. After three years of improving the system and investing in new technology and worker training, the system has reduced SSIs to 74 percent.
The Federal Aviation Administration (FAA) and defense contractor CACI have successfully piloted a system called SkyTracker that can detect drones flying in commercial airspace around airports and locate their operators. SkyTracker can sense radio signals used by drones to communicate with their controller and pinpoint the location of the drone operator for authorities without interfering with airport operations. FAA developed the system to help protect airplanes from getting too close to hobby drones that could damage a plane’s engines or cause other mechanical problems.
The President’s Council of Advisors on Science and Technology (PCAST) has submitted to President Obama its “Technology and the Future of Cities” report, exploring the potential for information and communication technologies to improve cities and identifying key opportunities for the U.S. government to accelerate their deployment. The report includes several recommendations for large-scale, coordinated action, including an interagency initiative led by the Department of Commerce to coordinate and support smart city efforts, encouraging the Department of Housing and Urban Development to embrace connected technologies to better serve low-income communities, and increasing federal funding for smart city initiatives.
Researchers at Stanford University have developed a machine learning system for mapping poverty around the world that compares light levels at night, which indicate electrification, with daytime images of infrastructure and the landscape—both provided by satellite imaging—to analyze economic activity. Electrification has long been a reliable indicator of poverty levels, but by combining daytime imagery, the researchers can gain more nuanced insights into poverty levels in specific locations.
Facebook has created population distribution maps of 20 countries where Internet penetration has not reached many rural populations, covering 2 billion people with a level of detail and accuracy that surpasses existing population maps. Facebook started with the most granular world population data set available, developed by Columbia University, merged it with high-resolution satellite imagery, and then applied an artificial neural network to distribute population estimates for an area based on buildings it could identify in the images. The maps, which Facebook will release this summer, will provide urban planning efforts, disaster response, and a wide variety of other applications with crucial information about just where people live.
Google-owned robotics company Boston Dynamics has released a demonstration of the second generation of its two-legged robot, named Atlas 2, that relies on an array of sensors and algorithms to navigate the environment, move objects, and respond to sudden changes, such as being pushed over or having a package hit out of its robotic arms. Atlas 2 uses light detection and ranging (LIDAR) technology to “see” the environment and relies on sensors throughout its body and legs to maintain its balance in varied terrain.
Image: Julien Willem.