This week’s list of data news highlights covers January 23, 2021 – January 29, 2021 and includes articles about assessing student engagement in virtual classrooms and identifying autism in children from biomarkers in their mothers.
1. Improving the Capabilities of Robots Using a Venus Flytrap
Researchers at the Nanyang Technological University in Singapore have explored new approaches to improve the capabilities and accuracy of robots by looking to Venus flytraps, plants that shut tight when stimulated by the movement of insects. The Venus flytrap waits for insects that land inside it to move two or three times to ensure it is only expending energy on insects that are alive. Researchers are trying to incorporate such network-like communications in robots. For instance by integrating portions of the flytrap into a robotic arm, researchers used a cellphone to transmit an electric pulse that triggered the flytrap to grasp a tiny piece of wire.
2. Assessing Student Engagement in Virtual Classrooms
Researchers at University of Tübingen and Leibniz Institute für Wissensmedien in Germany have collaborated with the University of Colorado Boulder to develop a neural network that estimates student engagement from video footage. The team trained one neural network to gauge a students’s attention from videos by identifying whether their heads were pointed towards or away from a web camera, and trained another to analyze their facial expressions.
Researchers at the University of East Anglia in England have designed a monitoring device to help users control their nicotine cravings and reduce the amount they smoke. The device screws on to tank-based e-cigarettes and measures the number and duration of a user’s vapes, the time between each drag, and the power used per puff. The device then sends the recorded data is then sent to an app that uses AI to create personalized portfolios of a user’s vaping patterns and offers suggestions to help them quit smoking.
4. Identifying Infrastructure Gaps in Electric Vehicle Charging Stations
Researchers at the Georgia Institute of Technology have developed an AI system to read electrical vehicle (EV) charging station reviews and predict places where there are an insufficient number of out-of-service stations. Researchers trained the system to parse through 12,720 reviews left on charge station locator apps and classify them into eight categories: functionality, availability, cost, location, dealership, user interaction, service time, and range anxiety. When tested, the system achieved a 91 percent accuracy rate in parsing through reviews.
5. Identifying Autism in Children from Biomarkers in their Mothers
Researchers from the University of California, Davis have identified biomarkers in mothers that indicate the likelihood and severity of autism in children. Pregnant women have immune proteins that can react with their growing fetus’s brain after its development. The researchers used machine learning to analyze plasma samples from 792 mothers, 450 of which had children with autism, and detected three types of reactions commonly associated with 20 percent of autism diagnoses.
6. Reducing Knee Pain Disparities in Arthritis Patients with AI
Researchers from Stanford, Harvard, the University of Chicago, and the University of California, Berkeley have developed a computer vision tool that identifies patterns of pixels in x-ray images that correlate with pain. When tested, the algorithm predicted pain levels that correlated more closely with the actual pain patients experienced than the scores radiologists assigned, particularly for Black patients.
7. Using AI to Safeguard Water Quality
The city of Newark, New Jersey is using AI to safeguard water quality. One AI system uses data from the city’s current and historic water monitoring data and factors in environmental elements, such as time of year, temperature, and precipitation, to predict whether adjustments are needed and what they should be.
8. Analyzing Data from Arctic Ice Sheets
Researchers at the University of Maryland have developed an AI algorithm that analyzes data from ice sheets in the Arctic and Antarctic. Previously, researchers would have to manually analyze large volumes of polar ice, snow, and soil measurements. But researchers have trained an algorithm to automate this process by identifying objects and patterns in the data, such as the level of snow accumulation in one area. With the algorithm, scientists can gain more insight into melting ice trends caused by climate change.
9. Estimating Human Emotional and Excitement Levels
Researchers at Samsung and Imperial College London have developed a neural-network that analyzes images of human faces taken in everyday settings to estimate their levels of excitement and whether they were displaying positive or negative emotions. The team trained the system to analyze facial expressions and their intensities based on how a person’s lips, nose, and eyes were positioned on a set of annotated images. When tested on two different image datasets, the system performed as well as expert human annotators.
10. Assessing the Risk of COVID-19 Patients for Respiratory Failure
Clinicians at the Feinstein Institutes for Medical Research in New York have developed an AI system that assesses the risk of COVID-19 patients experiencing respiratory failure within 48 hours. The clinicians trained the tool to extract and analyze vital signs, lab test results, and demographics from the electronic health records of 11,525 patients, 933 of which were placed on ventilation within 48 hours of admission. The tool used this analysis to give patients a risk score and when tested, had an accuracy rate of 92 percent.
Image credit: Annie Spratt
