This week’s list of data news highlights covers February 1-7, 2020, and includes articles about finding biological links in schizophrenia and an AI system that can automatically deliver sports updates in a video format.
Researchers from Jigsaw—a subsidiary of Google—and several universities have developed Assembler, a suite of tools that can detect manipulated images. One of the tools uses machine learning to differentiate between images of real people and deepfake images, and other tools can detect anomalies in color patterns and areas in images that show signs of copying and pasting. More than a dozen news and fact-checking organizations are testing the tool.
Researchers from Rice University have developed a system that uses AI to accurately predict extreme weather. The researchers trained the deep learning model on surface temperature and air pressure data, which they labeled according to the presence or lack of extreme weather events such as heatwaves and cold spells. The model can forecast severe weather for five-day forecasts with 85 percent accuracy.
Researchers led by an individual from the University of Washington have performed the first genetic analysis of schizophrenia in an African population, and the results suggest that schizophrenia may have biological origins. The study sequenced the genes of African individuals with and without schizophrenia and compared the results to those from a study of Swedish individuals, finding that individuals with schizophrenia in each population had mutations in similar genes. The mutations were often in genes important for brain development.
The Australian state of Victoria is implementing a system that uses AI to find keywords and patterns in the audio of emergency calls that indicate a person is suffering from cardiac arrest. In such instances, the system will alert the call taker to dispatch a high-priority ambulance and tell potential bystanders how to provide CPR or use a defibrillator.
Researchers from the University of Eastern Finland have developed a machine learning model that can identify different sleep stages as accurately as an experienced physician. Identifying sleep stages is essential to diagnosing sleep disorders, and the researchers trained the system on data from both healthy individuals and individuals suffering from sleep apnea. The model analyzed the electrical activity in individuals’ brains to identify the correct sleep stage with roughly 84 percent accuracy.
Researchers from Babylon Health, a digital healthcare provider based in the UK, have developed a system that can identify causal relationships between variables across datasets. The algorithm uses techniques from quantum cryptography, in which a mathematical formula can prove nobody has intercepted information, to discern if one variable causes another. The system accurately determined that the size and texture of breast tumors do not cause one another but that a tumor’s malignancy caused both.
Reuters and Synthesia, an AI startup based in London, have developed an AI system that uses pre-recorded footage of a news presenter to create new video reports about English Premier League football matches. The system combines real-time photography and reporting from matches to create a news script and uses a generative adversarial network to produce a digital twin of Reuters‘ sports editor that speaks the script.
Researchers from McGill University have developed a machine learning algorithm that can diagnose the severity of neurogenerative diseases in patients. The researchers evaluated the algorithm on gene expression data from cells in the prefrontal cortex and blood samples, finding that the algorithm’s assessment of the difference in gene expression in patients with and without dementia had a significant association with the severity of their dementia. The algorithm could help doctors evaluate patients and prescribe appropriate therapies.
The U.S. Department of Transportation has granted its first autonomous vehicle exemption to Silicon Valley startup Nuro, allowing the firm to test vehicles that lack human controls—such as steering wheels, pedals, and sideview mirrors—on public roads. Nuro will test R2, an autonomous delivery vehicle that operates at speeds below 25 miles per hour. Customers can access their food from the vehicle by inputting a code on a touchscreen.
Researchers from Scripps Research Translational Institute in San Diego analyzed resting heart rate data from 92,000 FitBit-wearers to find that the average resting heart rate for different individuals can range between 40 and 109 beats per minute. The researchers also found a seasonal trend: Subjects’ resting heart rates were lowest in the middle of the summer and highest at the beginning of the year. In addition, individuals who got a better night’s sleep had lower resting heart rates.