This week’s list of data news highlights covers August 15 – August 21, 2020, and includes articles about classifying galaxies using deep learning and using AI to quicken MRI scans
Ancestry, a U.S.-based genealogy company, has collected DNA data from 750,000 participants in order to investigate the link between genes and COVID-19 susceptibility. Researchers discovered a region on chromosome 1 that increases the likelihood of males contracting the virus by 44 percent. This discovery may explain why males who contract the virus experience more severe symptoms than their female counterparts.
Cedars-Sinai, a nonprofit hospital in California, has created a COVID-19 forecasting tool using machine learning. The tool can predict staffing needs and hospital capacities, as well as gauge personal protective equipment (PPE) supply levels. Furthermore, the tool can predict patients’ length of stay and their likelihood for readmission.
Researchers at Facebook have collaborated with radiologists at New York University to create fastMRI, a model that uses AI to make MRI scans four times faster. First, the model uses high-resolution images from a patient’s MRI scan to create an overall outline. Then, as the MRI scan progresses, fastMRI uses the lower-resolution images it receives to fill in gaps and produce a completed final scan. Unlike previous models that required humans to check the final scan, an AI system checks fastMRI to ensure the final MRI scan is consistent with MRI physics.
The Kuikuro Indigenous Association of Upper Xingu, an indigenous group in Brazil, has developed a COVID-19 community contact tracing app that collects data on travelers from each household, including their date of departure, the villages they traveled to and from, and whether or not they presented symptoms within the last 14 days. Since June, the Kuikuro village has only 210 confirmed cases of COVID-19, which is significantly lower than the 26,000 confirmed cases among Brazil’s other indigenous groups.
A team of researchers from the Polytechnic University of Catalonia in Spain, known as BarcelonaTech, have developed an acoustic elephant detector to prevent elephant-train collisions in India. The team installed recorders and cameras on the Siliguri-Jalpaiguri railway and used machine learning to analyze different animal sounds and images to correctly identify elephants. Because previous collisions occurred largely at night, the team also installed thermal sensors to enhance detection. The detector first identifies an elephant by sound, image, or body temperature, and then alerts train conductors via Wi-Fi.
Researchers at Stanford University have created an AI software called Vid2Player, which can generate realistic tennis matches between professional players. To increase the realism of matches, researchers used video footage of professional tennis players to create sprites, which are characters based on videos of real people. Additionally, users can interact with the software to change the course of a match, such as by altering the way a player aims and hits the ball.
Spotify is using AI technology to create a personalized music experience for the Grammy-award winning Canadian musician The Weeknd’s After Hours album on a microsite. His After Hours album landed 2 top hits on the Billboard 100, and is considered a front-runner for this year’s Album of the Year Grammy award. When users enter the microsite, an avatar of The Weeknd appears on screen and addresses each user by name. Using the user’s streaming history, The Weeknd’s avatar provides a unique virtual concert experience by singing the user’s most played songs from his album.
Scientists from the National Astronomical Observatory of Japan have identified more than 560,000 galaxies using a deep-learning algorithm. Scientists trained the algorithm using images of spiral-patterned galaxies similar to the Milky Way. When used on the test set, the algorithm accurately classified 95.7 percent of galaxies.
The city of London, Ontario in Canada has developed an AI tool called Chronic Homelessness Artificial Intelligence (CHAI) that predicts the likelihood of a person seeking shelter services or becoming homeless within the next six months. CHAI uses 21 million data points from the city’s real-time homelessness database and weighs factors like age, gender, and shelter history to make its predictions. CHAI offers explanations for its predictions and has a 93 percent accuracy rating.
NoTraffic, a California-based startup focused on traffic management, has created a traffic grid management system to improve traffic flow and reduce vehicle delays. Instead of a typical timer-based model, the system autonomously coordinates traffic lights based on demand. At each intersection, installed AI sensors differentiate between pedestrians and vehicle types. Using this data, the system calculates the number of vehicles present and changes traffic lights accordingly. When tested, the system reduced vehicle delay time by 40 percent and the city of Phoenix, Arizona is piloting the technology.