This week’s list of data news highlights covers May 16-22, 2020, and includes articles about predicting the seasonality of COVID-19 and using autonomous drones to deliver defibrillators in emergency situations.
Researchers from Lawrence Berkeley National Laboratory are using machine learning to predict COVID-19’s seasonal cycle. The researchers are analyzing health, demographic, climate, and mobility data. The researchers’ goal is to predict how environmental factors affect the transmission of the disease for each county in the United States, which could help policymakers make decisions about loosening social distancing restrictions.
Researchers from Humboldt University of Berlin have used declassified U.S. satellite images to reveal a significant decline in Kazakhstan’s biodiversity. The researchers compared images from 1968, 1969, 1999, and 2017, finding that the number of burrows from the squirrel-like marmot declined 14 percent between 1968 and 2017. In areas that became farmland, the number of burrows decreased by 60 percent.
Researchers from the University of Maryland have developed a system that helps autonomous robots decide how much space to provide a human based on their perception of the individual’s emotional state. The researchers developed the system by extracting the skeletal gaits of people walking in videos, which they had labeled into emotional categories, such as angry, sad, happy, and neutral. Deep learning algorithms then learned to associate specific gaits with particular emotions.
Facebook has developed a computer vision system that can identify products in photos and automatically generate descriptions. Facebook trained the system on millions of images from seven datasets, including Facebook Marketplace. The system, which can identify occluded objects, such as a shirt beneath a jacket, can also predict likely search queries for the product.
Researchers led by an individual from the University of Melbourne have sequenced the genomes of 75 percent of the coronavirus cases in Victoria, Australia. The researchers identified 76 distinct genomic clusters, stemming from areas such as social venues, healthcare facilities, and cruise ships. The researchers also found that transmission decreased significantly following travel restrictions and the implementation of social distancing.
Sattva MedTech, an Indian medical startup, has developed an AI-enabled device that can diagnose fetal distress, which occurs when a fetus has not received enough oxygen. The tool analyzes electrocardiogram data, detecting the condition with 93 percent accuracy.
Everdrone, a Swedish company that creates software for autonomous drones, is using autonomous drones to deliver defibrillators to the scenes of cardiac arrests in Sweden. The survival rate for out-of-hospital cardiac arrests is roughly 10 percent. The service, which is available to more than 80,000 residents, is part of a clinical study that will examine the effectiveness of using drones for emergency operations.
Facebook has developed an AI-enabled tool to flag potential scammers and imposters on Facebook Messenger. The tool uses behavioral signals, such as an account that uses a name similar to the user’s friend or an account that sends a significant number of friend requests and messages to minors, to warn users. The tool does not read the content of messages.
The county of Miami-Dade, Florida, has partnered with BioBot, a wastewater epidemiology firm based in Massachusetts, to track the spread of COVID-19. The firm analyzes samples of untreated waste to detect signs of COVID-19 and estimate the county’s infection rate. The county plans to use the data to predict disease hotspots and new rounds of infections.
Researchers from Waymo, the autonomous vehicle subsidiary of Alphabet, have developed an AI-enabled system that can create camera images to train autonomous vehicles in simulation. The system uses real-world data from autonomous vehicles, such as LiDAR and camera data, to generate 3D data for all objects in a scene. The system, which renders a simulated scene from multiple angles, uses a generative adversarial network to create images that are difficult to discern from real images.