This week’s list of data news highlights covers October 20-26, 2018, and includes articles about an AI tool that can create labeled training data and a bot that can patch code as well as humans.
Nvidia and Scripps Research Translation Institute, a non-profit research organization, have partnered to develop AI-powered tools that leverage digital health and genomic data. One of the partnership’s first projects will focus on using data from wearable devices, such as smartwatches, to predict atrial fibrillation, an irregular heartbeat that increases the risk of stroke. The organizations also plan to use AI to improve the detection of mutations in genomic data.
Researchers at the University of Texas at Austin have developed a method for sequencing proteins that is much more sensitive than existing methods. Traditional protein sequencing methods rely on a technique called mass spectrometry, which can only detect a protein if there are at least a million copies of it, and can only detect a few thousand distinct protein types in a sample. The researchers’ method, called single molecule fleurosequencing, is similar to DNA sequencing in that it can identify single protein molecules in a sample, which could enable healthcare researchers to better detect biomarkers for diseases.
Google has a created a tool called Fluid Annotation that uses AI to automatically generate labels for objects in image datasets. Labeling objects in computer vision datasets often requires human annotators, which makes the process of developing training datasets for AI time intensive. Fluid Annotation automatically suggests potential labels which humans can then verify.
Aircraft manufacturer Airbus has developed a series of data-driven tools to help airlines better manage their fleets and repairs, which could reduce costly flight delays. Airbus’ Scheduled Maintenance Optimizer platform uses algorithms to recommend maintenance schedules for fleets with over 200 aircraft. And Airbus’ Skywise tool pulls aviation data from a variety of industry sources, such as plane sensor data, flight schedules, and pilot reports, into an analytics platform to improve industrial operations.
A group of U.S. and U.K. university researchers have developed an AI tool called AI Clinician to help doctors treat sepsis, a blood infection that kills nearly 270,000 U.S. patients every year. The researchers trained the tool on data from 96,000 intensive care unit patients, which included information about patients’ demographics, vital signs, and fluids or drugs they had received. While it is difficult for doctors to identify when they should administer fluids and drugs to patients to treat sepsis, and how much to give them, researchers found that patients whose actual treatment matched AI Clinicians had the lowest mortality rates.
Researchers at the Swedish KTH Royal Institute of Technology, have created an AI bot called Repairnator that can fix software bugs in code faster than humans. The bot constantly monitors the code sharing website GitHub for bugs and attempts to produce patches before human developers can. While researchers have created bots before to patch bugs, the bots usually produce poorly written code that humans would not accept. Repairnator, however, has written five patches that human engineers have already accepted.
The Massachusetts Institute of Technology has completed an experiment called Moral Machine that crowdsourced data about how people from different cultures would respond to variations of the trolly problem, a thought experiment in which a person must decide whether an out-of-control trolly should hit one person or five people, which plagues popular debates about autonomous vehicles. The data, sourced from 40 million responses from people in 233 countries and territories, demonstrated significant divergence in ethical decisions depending on factors such as culture and economics. For example, respondents from Taiwan and China, where respect for elders is held in high regard, were more likely to think an autonomous car should spare the lives of the elderly, while respondents from France were more likely to think it should spare the lives of the young.
The Massachusetts Institute of Technology has launched a food sustainability program called the Open Agriculture Initiative to identify the ideal growing conditions and flavors for different plants with the help of AI. For example, the lab is working with Ferrero, the maker of Nutella, to identify the best growing conditions for hazelnuts by creating an indoor farm in which AI controls variables such as temperature, humidity, and pH levels. Ferrero plans on comparing the data with climate and soil data around the world to choose new areas to farm.
Several companies are deploying innovative, low-cost screening technologies for breast cancer in India in an attempt to reduce the country’s high mortality rate for the disease. Mortality rates are because healthcare in India is expensive and inaccessible which makes early detection unlikely. A company called Niramai is experimenting with using machine learning to analyze scans from a tripod-mounted thermal camera to detect signs of tumors for just $20 per test. Another company called MammoAlert will distribute lab-in-a-box kits to collect blood samples, analyze them with AI, and send results via smartphone within 30 minutes. A third tool, developed by iBreastExam, uses a handheld scanner with 16 sensors that can measure tissue stiffness and relay results to a smartphone app with a high degree of accuracy.
New York startup Clear has developed a service that allows customers to use their biometric data, including their fingerprints, irises, and faces, to more quickly pass through security at airports and stadiums. For an annual fee, the company, which was spun out of airport security screening firm Verified Identity Pass, offers users the ability to register their biometric data and then pass through security with just a fingerprint scan to verify their identity. Though similar services exist at U.S. airports, Clear is expanding the model to other venues, including sports and entertainment arenas where users can also pay for concessions with their fingerprints. Clear was able to reduce security passage times for users at Seattle Mariners games from five to ten minutes to just one minute.