This week’s list of data news highlights covers January 21-27, 2017 and includes articles about an AI program winning a poker tournament and a new effort to collect accurate data about hate crimes.
Medical specialists at Austin Health Liver Transplant Unit in Melbourne, Australia, have developed a machine learning algorithm that can match liver donors and transplant recipients. The specialists modeled the algorithm after the matching algorithm used by online dating website eHarmony, using 25 different medical factors, such as blood type and age to find ideal matches. In a test on historical transplant data, the algorithm was able to match donors with 84 percent accuracy, compared to 68 percent accuracy rate of the traditional method of using humans to identify matches.
The government in Louisville, Kentucky, is working with volunteers and data-focused civil society groups to develop a suite of municipal services that can interact with smart devices in residents’ homes. So far, the city has developed a program that will automatically adjust the hue of a smart lightbulb to represent the city’s air quality, as well as a services for Amazon’s smart home device Alexa that allows users to access information about neighborhood trash removal and news from the mayor using voice commands. Separately, the city has also installed Internet-connected smoke detectors in vacant properties that can automatically alert the fire department if it detects a fire.
Australian company Fluid Management Technology (FMT) has developed a system called SmartFill that uses connected devices to monitor fuel consumption for vehicle fleets to make it more difficult for people to steal fuel. SmartFill uses sensors to authenticate which driver and vehicle are attempting to refuel and record how much fuel is dispensed, while software analyzes data from vehicle odometers and cargo lists to calculate how much fuel a particular vehicle should need. If the system detects a discrepancy, it will notify a manager that someone might be stealing fuel. FMT also allows customers to use SmartFill data to claim rebates on Australia’s diesel fuel excises taxes.
A poker-playing AI system developed by researchers at Carnegie Mellon University named Libratus is consistently beating human players in a poker tournament in Pittsburgh, Pennsylvania. Poker is challenging for computer systems to play at high levels due to the fact that a system cannot know all of the information needed to make the best possible decision, as opponents hands are hidden. Libratus has played 67,000 games of poker, earned $701,242 worth of chips, and is currently in first place in the tournament.
Journalism nonprofit ProPublica, in partnership with civil rights organizations and other journalism organizations, has launched an initiative called Documenting Hate that will pool user-submitted datasets on hate crimes to supplement federal statistics and provide journalists with useful data. Local police departments do not report data on hate crimes to the federal government in a consistent manner, making it difficult for journalists and members of the public to understand the prevalence of these incidents. Documenting Hate allows users to anonymously report hate crimes and ProPublica is working with mapping organizations and local civil rights groups to collect and analyze hate crime data.
British startup Jukedeck has developed AI software that can automatically generate unique music based on a customer’s’ specifications. Jukedeck trained its software on hundreds of different scores so it could learn relationships between different aspects of music, such as the likelihood of a particular chord progression, to allow it to generate natural-sounding music. Jukedeck then uses a separate neural network to produce the music. Jukedeck licenses its AI-developed music to customers for just $21.99, which is substantially less than it would cost for a company to license music written and performed by humans.
Stanford University researchers have taught an algorithm to analyze images of skin lesions and identify skin cancers as accurately as human experts, if not better. The researchers trained deep learning algorithm designed by Google for classifying images on a database of 130,000 images of moles, rashes, and lesions, and they. In tests, their algorithm could correctly identify 96 percent of malignant lesions and 90 percent of benign lesions.
Endgame, a cybersecurity firm, has developed a virtual assistant called “Artemis” that can provide computer analysts information about security threats. The virtual assistant is a chatbot that can answer questions such as “what is suspicious about my network today?” The software uses machine learning and natural language processing to detect malicious activity on computer networks.
The science and education nonprofit Chan Zuckerberg Initiative has announced that it is acquiring a search engine called Meta that uses AI to help researchers find science research papers and that it will make Meta free for the public to use. Meta creates indexes of research repositories by crawling the Internet to identify and rank authors and analyze other research that they cite. These indexes make it easier for researchers to find specific papers they are looking for as well as other papers that could be relevant. By making Meta free to use, the Chan Zuckerberg Initiative will enable schools, students, scientists, and groups that sponsor research to easily parse through huge amounts of scientific research
Researchers at Boston Children’s Hospital and the Dana-Farber Cancer Institute have found that analyzing tumor genomes could substantially improve how doctors diagnose and treat pediatric brain cancer. The researchers analyzed 203 patients’ tumor genomes and were able to identify genetic abnormalities in 56 percent of them that could allow doctors to make more personalized treatment decisions. Genomic analysis can be particularly useful for pediatric brain tumors as there are no U.S. Food and Drug Administration-approved drugs designed specifically for pediatric brain cancers, making additional information about what kind of treatments might be effective for specific patients more valuable for doctors.
Image: U.S. Navy.