This week’s list of data news highlights covers November 9-November 15, 2019, and includes articles about AI helping clean up sewers and using satellite data to create a map of human settlements.
Consumer-genetics firm Ancestry used machine learning algorithms to automatically extract data from more than 500 million obituaries in newspapers dating back to 1690. The firm’s system used one algorithm to detect obituary content in newspapers and another algorithm to identify key facts such as the name of an individual’s spouse, children, and birthplace. The firm trained the algorithms on thousands of newspaper obituaries people annotated, and the system can help customers discover information that helps them build family trees.
Water services companies are using computer vision and other forms of AI to help repair and remove conglomerations of fat, wet wipes, and other non-biodegradable items from sewers. For example, DC Water, which operates in Washington, DC, uses software to automatically analyze images from robots to identify buildups of fat, invading plant roots, and broken pipes. The software helps reduce the costs of inspecting sewers by as much as 75 percent.
Royal Dutch Shell, the oil and gas firm, is using sensors and machine learning at Europe’s largest oil refinery to detect potential failures in control valves 75 days in advance. The refinery has 50,000 sensors, which generate 100,000 measurements a minute that a machine learning model analyzes. The model allows workers to perform maintenance or adjust operating conditions prior to significant issues.
Researchers from Microsoft and SRL Diagnostics, an Indian diagnostics laboratory firm, have developed software that uses AI to screen for cervical cancer by analyzing smear images. The researchers trained the software’s AI model on thousands of annotated images of cervical smears. The software can accurately identify images that display abnormalities, which can help doctors spend time on cases that justify concern.
IBM has created a new weather forecasting model that leverages big data and a new supercomputer to create forecasts for areas as small as two miles wide, compared to other weather models that typically provide forecasts for six to nine-mile areas. The model, which can use data from airplane sensors and smartphones to provide more accurate temperature and pressure data, will process forecasts every hour. Many other global weather models update every six to twelve hours.
Researchers from Geisinger, a healthcare provider based in Pennsylvania, have developed a neural network that can predict which patients are likely to develop irregular heartbeats by analyzing electrocardiograms (ECG). Researchers trained the neural network using 30 years of ECG results, finding that the networks could predict which patients would develop irregular heartbeats even when doctors interpreted their current heartbeat as normal. The researchers also found that the network could predict a patient’s risk of death more accurately than doctors.
Researchers from the German Aerospace Center have used radar and optical images from multiple satellites to create a global, open-source map of human settlements that has a resolution of 10 meters. The researchers used a supercomputer and machine learning algorithm to spot human settlements in over 300,000 images of the Earth’s surface. The map and its underlying data can help demographers and sociologists better understand human settlements, particularly in remote areas.
Researchers from the University of Bristol and the University of Bath in the UK have developed a pacemaker that uses a neural network to increase the amount of blood rats’ hearts can pump by 20 percent compared to other pacemakers. Traditional pacemakers do not take into account the natural variation in heartbeats that happens as an animal inhales and exhales, but the researchers’ pacemaker analyzes the electrical activity from the rats’ diaphragm muscles to prompt the heart to beat in sync with breathing.
LG has developed an AI-enabled system that analyzes data from home appliances to alert customers about potential issues before they occur. LG trained the system’s algorithms to detect conditions that lead to mechanical problems, including using too much laundry detergent, which can cause leaks in washing machines. The system alerts LG customers through an app to the potential breakdowns and recommends ways to prevent them.
Researchers from Yonsei University in South Korea have shown that AI-enabled software can help clinicians more accurately analyze X-rays for lung cancer. The researchers trained and tested commercially available deep learning software on more than 19,000 X-ray images. The researchers found that the software, which they trained to identify potentially malignant lung nodules, improves the ability of radiologists to accurately identify existing lung cancer from 65 percent to 70 percent.