This week’s list of data news highlights covers May 9-15, 2020, and includes articles about evaluating the validity of COVID-19 research and using AI to detect hate speech.
The U.S. National Institutes of Health and Nvidia have developed an AI system that can detect signs of COVID-19 in lung scans. The groups developed the system using a dataset of 3,000 computed tomography (CT) images from patients in the United States, Italy, China, and Japan. More than 2,000 of these images are of individuals who tested positive for COVID-19. The system also creates a heat map that highlights where the virus is present.
The University of Pennsylvania is leading a group of 30 healthcare organizations that are using federated learning to develop an AI system that can detect the boundaries of brain tumors in images. Federated learning allows the organizations to train the same AI model on data from multiple locations without having to pool the data at a central location. This process enables the group to train the system on more data while overcoming the regulatory and privacy restrictions associated with sharing health data.
Researchers from Northwestern University have developed a tool that can assess the level of confidence the authors of COVID-19 research papers have in their findings. The tools model uses natural language processing to evaluate researchers’ confidence level, which acts as a proxy for determining if the research is likely replicable. The researchers found that several words appearing together, including “subjects,” “experiment,” and “significance,” conveyed confidence.
Amazon has developed a machine learning system that can act as an internal search engine for firms. The tool allows users to ask questions, such as “How do I connect to my VPN?” instead of using keywords. The tool uses natural language processing to return answers, which range from snippets from documents to FAQs to entire documents.
Nearly 89 percent of all the hate speech Facebook removed this quarter was detected by AI, increasing from 80 percent the previous quarter. Facebook’s improvements come from using natural language models that can better understand nuances in posts and systems that can analyze content that consists of a combination of text and images, such as a hateful meme.
South Africa has redesigned a system it uses to track poaching to track the spread of coronavirus. The redesigned system collects data from diagnostic laboratory tests, screening interviews from community workers, and demographic information about the spread of the nation’s population. If an individual tests positive for COVID-19, health officials receive an alert with the person’s address information and begin contact tracing.
Researchers from Monash University in Australia have developed an AI-enabled system that can predict if and when patients with chronic liver disease and heart failure will return to a hospital. The researchers developed the system using 14,000 patient records and data on 327,000 hospital readmissions. The system could help doctors decide when to deliver timely interventions to prevent hospitalization.
Researchers from the U.S. National Institutes of Health have used lasers and movie recordings of people talking to determine that droplets from individuals talking can stay in the air for up to 14 minutes. The researchers had individuals repeat the phrase “stay healthy” and used lasers to visualize the droplets. The researchers analyzed the recordings frame by frame to count the number of droplets over time.
23andMe is offering free genetic tests to 10,000 individuals who have been hospitalized due to COVID-19 to determine if genes play a role in individuals’ symptoms. The company will combine the data with data from 400,000 of its customers—6,000 of whom tested positive for COVID-19—that responded to an online survey about their symptoms.
Uber is using a computer vision algorithm to verify that its drivers have face coverings. Before driving, drivers will take a selfie in Uber’s app. The algorithm will then detect the presence or absence of a mask without processing any biometric information or comparing the selfies to photos in Uber’s database of driver photos.