This week’s list of data news highlights covers October 26-November 1, 2019, and includes articles about analyzing social media posts to identify individuals who may commit violence and the use of near-infrared and visible spectrum images from satellites to track a fire in real-time.
Researchers from MIT have used a supercomputing system to create a model of what the world’s web traffic looks like on an average day, which can help researchers identify anomalies that stem from malicious cyber activity. The researchers used a neural network and the 10,000 processors of the MIT SuperCloud, which combines disparate computing resources, to analyze an Internet traffic dataset comprising 50 billion data packets. The model revealed the links between source and destination data points, such as between consumers and popular applications.
Researchers from the Sidney Kimmel Cancer Center in Philadelphia have developed a machine learning algorithm that can identify if thyroid nodules, which are small bumps in the thyroid gland, are malignant with 77 percent accuracy. The researchers trained the algorithm on ultrasound images from 121 patients who underwent a biopsy. The algorithm correctly identified 90 percent of malignant nodules and 74 percent of the nodules it identified as benign were benign.
Researchers from MIT and the University of Massachusetts, Lowell, used 185,000 words posted online by individuals who became mass shooters or terrorists and 50,000 words published online by people who did not commit violent acts to identify statistical differences in word choice. The researchers found that individuals who became violent more often used emotional and targeted words such as “hate,” “you,” and “they” while being less likely to use words about the external world, such as “people” and “world.”
Canadian Internet of Things firm SensorUp has developed a wearable device for firefighters that tracks an individual’s exposure to chemicals, their location, and if they have fallen down. The device also tracks firefighter’s vital signs such as temperature and heart rate. The system aggregates the data into a dashboard that updates in real-time.
Researchers from Climate Central, a U.S. organization that analyzes the effects of climate change, developed an AI model that improves upon previous predictions for the number of individuals living on land threatened by rising sea levels. Other calculations of individuals at risk use satellite data to estimate land elevations, which has led to incorrect elevation estimates because objects such as skyscrapers in dense cities can be mistaken for the ground. The researchers trained their model on several datasets, including maps that used laser light from planes to create more accurate elevation maps. The researchers’ model predicts that rising sea levels threaten as many as 250 million people, compared to previous estimates of 65 million.
Researchers from Google have developed a system that uses machine learning to predict what a molecule smells like by analyzing its structure. The researchers trained the system on datasets of molecules that experts labeled with one or more odor descriptors. The system could help in the development of synthetic fragrances.
Researchers from MIT have created a system that can analyze shadows to predict if an object will come around a corner, which could be useful for self-driving cars and robots. The system uses optical cameras and computer vision to track changes in the strength and intensity of light, allowing it to identify stationary objects and predict a moving object’s path. The system makes its predictions half a second faster compared to similar systems that use LIDAR.
Google is using near-infrared and visible spectrum images from satellites to create near-real-time maps of how the Kincade Fire in Northern California is spreading. Google can update the map every 5-20 minutes, helping individuals quickly assess their situation in an emergency.
Cape Analytics, a U.S. firm that uses AI to provide information to home insurance providers, used computer vision to find that less than two percent of U.S. homes have solar panels. The firm analyzed 38 million properties in 21 major metropolitan areas, finding San Diego has the most solar panels per 100,000 homes with more than 10,000.
DeepMind has developed AI agents that are capable of beating 99.8 percent of all human players at the video game StarCraft II. While DeepMind’s agents have defeated top professional players before, DeepMind provided the agents advantages such as being able to see the entire game map. In this new round of competition, DeepMind only allowed the agents to see a portion of the map a human would see and limited the number of mouse clicks agents could make to align with standard human movement.