This week’s list of data news highlights covers January 28 – February 3, 2017 and includes articles about how analytics can reduce dropout rates and a system that can help patients with locked-in syndrome communicate.
An engineering professor at the University of Texas at Arlington has developed a prototype device that uses low-cost sensors to analyze a person’s breath to determine whether the individual has the flu. The sensors can detect the presence of nitric oxide and ammonia in breath, which are biomarkers of the flu virus, just like a breathalyzer can detect the presence of alcohol in the breath of someone who has been drinking. Also like a breathalyzer, the device is designed to be easy for the average person to use so patients do not need to rely on visiting a doctor to obtain a diagnosis.
Georgia State University has developed predictive analytics technology that can help administrators identify students at risk of dropping out to prompt preemptive intervention, and it is sharing this technology with 10 other universities participating in the University Innovation Alliance, which aims to improve the accessibility of higher education. The technology compares student data to historical data about student successes, and can alert administrators when a student exhibits the same signs as other students prior to them dropping out. The University Innovation Alliance has pledged to use this technology to ensure the graduation of an additional 68,000 students by 2025.
Researchers at the Massachusetts Institute of Technology have developed a system that uses wearable sensors and AI to analyze physiological data and speech patterns and classify emotional qualities of a conversation. A wearable wristband called the Samsung Simband captures high-resolution physiological data and a smartphone records a subject’s speech, while a neural network identifies patterns related to different emotions and can determine if the speech is positive, neutral, or negative. As such systems advance, they could be useful to people with Asperger’s syndrome, anxiety, or other social disorders that make it difficult for them to understand the emotional cues of others.
A team of Chinese computer scientists and ophthalmologists have developed a machine learning algorithm called CC-Cruiser capable of identifying the presence of congenital cataracts in humans with 90 percent accuracy. Congenital cataracts are responsible for 10 percent of all vision loss in children but is still a rare disease, which means doctors have a higher chance of overlooking the condition and making an incorrect diagnosis. CC-Cruiser can screen patients for the disease, identify the risk levels a patient with the disease faces, and recommend treatments.
Researchers at the Wyss Centre for Bio and Neuroengineering in Geneva have developed a system for analyzing biometric data from patients suffering from locked-in syndrome, which occurs when medical conditions such as amyotrophic lateral sclerosis (ALS) render a person nearly completely physically paralyzed but mentally intact, that allows the patients to communicate by answering yes-or-no questions. The system uses a cap outfitted with sensors that use a technique called near-infrared spectroscopy to monitor the brain’s electrical activity and blood flow. The researchers monitored how these patterns changed while asking patients simple yes-or-no questions to establish a baseline of what a “yes” or “no” looks like, and the system could detect yes-or-no answers to new questions with 70 percent accuracy.
Tax preparation firm H&R Block has partnered with IBM to use the Watson cognitive computing platform to analyze tax data, respond to customer questions, and identify ways customers could maximize their tax refunds. H&R Block trained Watson on examples of common questions it receives from customers about their taxes as well as thousands of documents about federal and state tax laws.
Cybersecurity startup ForAllSecure, which developed the AI system Mayhem that won the U.S. Defense Advanced Projects Research Agency’s (DARPA’s) Cyber Grand Challenge last year, has partnered with several firms to test using Mayhem to improve their cybersecurity. When a company discovers one of its devices has a cybersecurity vulnerability and is being exploited, it can take days or weeks to patch, and considering that many companies produce new devices and software updates many times a year, constantly monitoring for, fixing, and distributing patches for vulnerabilities can be very resource intensive. ForAllSecure will test how Mayhem can proactively identify these vulnerabilities, automatically detect as soon as one is exploited, and rapidly distribute patches to all potentially affected devices.
A team of developers from the University of Washington have released an online service called Access Map that allows users to examine the accessibility of different streets in Seattle, Washington, such as if a particular sidewalk has a sloped curb to allow wheelchair access, and generate walking directions based on a user’s desired level of accessibility. The team combined data from the U.S. Geological Survey, the Seattle Department of Transportation, and other sources to overlay accessibility data over existing mapping data of Seattle.
Facebook has launched a new search feature that allows users to search through their and their friends’ uploaded photos by searching for descriptions of their contents. The search feature uses computer vision technology to automatically identify the contents of images and populate search results with images that it thinks most closely match desired keywords.
After a 20 day poker tournament in Pittsburgh, Pennsylvania, an AI system named Libratus developed by researchers from Carnegie Mellon University has come out on top after playing a total 120,000 hands of Texas hold’em against humans, including playing against four of the world’s top players. Libratus’s winnings surpassed those of the leading human players by $1,766,250 in chips. Libratus was able to master poker, which traditionally has been difficult for computers due to the fact that a system cannot know all of the cards in play, by using the Pittsburgh Supercomputing Center’s Bridges supercomputer to analyze 2,600 terabytes of data and identify and improve weaknesses in its strategy on a daily basis.
Image: Shane T. McCoy