This week’s list of data news highlights covers November 30-December 6, 2019, and includes articles about using AI to reduce injuries in the NFL and using a quantum computer to route traffic.
Researchers from Google have developed a series of AI models that can analyze chest x-rays with the same accuracy as radiologists. The researchers trained the models using roughly 600,000 chest x-ray images, and the models can detect multiple conditions, including collapsed lungs. The researchers used a natural language processing model to analyze radiology reports and provide labels for the training images.
Facebook has used machine learning to remove 98 percent of terrorist videos and photos before any user sees them. The firm trained its neural networks to identify objects and label them with percentages of confidence using labeled videos from human reviewers and reports from users. This process helps the system determine if a video or photo displays problematic content, and if it does, Facebook automatically deletes that photo or video.
The U.S. Department of Veteran Affairs has launched the National Artificial Intelligence Institute to develop the VA’s AI research and development capabilities. The institute plans to leverage its extensive health data, including its genomic database containing 800,000 veterans’ data, to develop AI solutions that support veterans. The institute will focus on deep learning, explainable AI, privacy-preserving AI, and AI for multi-scale time series.
The National Football League (NFL) and Amazon Web Services (AWS) have partnered to use machine learning to reduce injuries. For example, the partnership will use computer vision to analyze footage to detect concussions and identify the levels of force that cause them. In addition, AWS and the NFL will develop a virtual version of an NFL player to model how different factors, such as playing surface, equipment, and environmental factors, affect player injuries and rehabilitation.
Researchers from Duke University, Harvard University, the University of Wisconsin, and Massachusetts General Hospital have developed a machine learning model that can predict which patients are likely to experience a seizure after an injury. The researchers trained the model on the data of thousands of patients, and the model provides a probability estimate of a patient having a severe seizure. The researchers used the model at the University of Wisconsin and Massachusetts General Hospital, finding that it reduced the time doctors used continuous electroencephalography monitoring, which is expensive, to detect if a patient is having seizures by nearly 64 percent.
Descartes Labs, a satellite imagery analytics company based in New Mexico, has developed a tool that automates the detection of wildfires by analyzing satellite imagery. The software uses AI to analyze images from U.S. government weather satellites every few minutes, allowing it to detect changes from previous images, such as the brightness of any pixels. The tool has detected more than 6,000 wildfires since July, and the tool has helped Descartes Labs report the fire to the relevant authorities as fast as nine minutes after satellite imagery captured the fire.
StreetLight Data, a U.S. analytics firm, has developed a platform that uses machine learning to analyze several types of big data to help cities improve their urban planning. The platform analyzes trillions of GPS, cellular, connected car, and IoT data points to learn the traffic patterns of cars, bikes, scooters, and pedestrians. This analysis helps cities make decisions such as where to create new bike paths or place electric vehicle chargers.
Volkswagen used a quantum computer to calculate the fastest travel routes for nine public transit buses. The quantum computer performed calculations that helped plot and update routes for the buses in near-real-time by predicting the flow of traffic, helping the buses avoid traffic jams before they occurred. Quantum computers could help connected navigation systems direct vehicles along different paths to avoid sending vehicles on the same road, which could create congestion.
Researchers from Washington University in St. Louis have developed a deep learning system that detected colorectal cancer tumors with 100 percent accuracy in a pilot test. The researchers trained and tested the system’s convolutional neural network on 26,000 images of 20 tumor areas, 16 benign areas, and 6 abnormal areas.
Facebook has developed AI-powered bots that can play Hanabi, a Japanese card game that requires teamwork. Hanabi participants work together to achieve the highest score possible by correctly arranging cards by color and number order. Each participant relies on cues from other players to arrange cards because each individual cannot see their own cards. Consequently, players must infer why other participants played certain cards. Facebook’s bots simulated hypothetical moves made by other players to determine the best card to play, and the bots consistently achieved higher scores than advanced human players.
Image: Tech. Sgt. Michael Holzworth