This week’s list of data news highlights covers October 1-7, 2016 and includes articles about predictive analytics software that can make airports more efficient and a new algorithm that can diagnose sick plants.
The U.S. Geological Survey (USGS) has expanded its network of connected environmental sensors in states affected by Hurricane Matthew to better inform storm tracking and disaster relief efforts. USGS has installed 305 new sensors throughout Florida, Georgia, North Carolina, and South Carolina that collect data about the storm, such as its direction, wind speed, and duration of the storm surge, and it can report this information in real-time from locations that would otherwise be inaccessible due to severe conditions. USGS is making data from these sensors publicly available on its Flood Viewer website, which visualizes data about major storms.
Mastercard has launched a smartphone app called Identity Check Mobile in 12 European countries that allows users to authorize payments with biometric data, including with a facial recognition algorithm that verifies the identify of a user based on a selfie. Identity Check Mobile can eliminate the need for users to manually enter passwords, and in testing, Mastercard says 92 percent of users preferred the app’s biometric analysis over using passwords. Mastercard plans to launch the app in the United States in 2017.
Researchers at Pennsylvania State University and the Swiss university École Polytechnique Fédérale de Lausanne have developed a deep learning system to analyze photographs of sick plants and diagnose the cause. The researchers trained their system on 50,000 photos of diseased plants so it could learn to differentiate between diseases and identify how they affect different plants. The system can identify 26 different diseases in 14 different plant species in images it has never analyzed before with over 99 percent accuracy, and the researchers plan to incorporate it into a smartphone app for farmers.
New York’s JFK airport has implemented predictive analytics software called Beontra that analyzes data on flight operations, such as plane delays and passenger counts, as well as data about the airport, such as how long it takes for people to walk through a terminal and pass through security, to predict passenger wait times. With these predictions, airlines and airport officials can make more informed decisions about scheduling and personnel management, as well as help passengers make more informed travel plans.
Researchers at Imperial College London have developed an AI model that combines machine learning with an AI technique called symbolic AI, which involves manual labeling of certain patterns and instructions for software to follow, to solve problems sooner than pure machine learning systems. Pure symbolic AI systems can be effective, but are very time consuming to develop and cannot apply their instructions to new situations, while pure machine learning systems are very adaptable, but must essentially start from scratch every time. The researcher’s hybrid model allows developers to give a system basic guidelines to learn from that allow the system to adapt what it has learned to new situations. In a test, the hybrid model was able to develop a strategy that won a simple board game 70 percent of the time after just 200 tries, while a deep learning algorithm developed by Google’s DeepMind took 1,000 tries to develop a strategy that was effective just 50 percent of the time.
Industrial robotics company Fanuc has partnered with Nvidia to develop industrial robotic systems that use machine learning to complete new tasks instead of having to be reprogrammed and given specific instructions every time they receive a new task. A centralized system will train models for completing tasks on operational data from Fanuc robots in a factory, which can help factory operators save the time and money that it takes for human experts to regularly reprogram their systems.
Researchers at the Massachusetts Institute of Technology have developed a method for capturing information from light that has passed through translucent material, such as human tissue. The researchers use a high-speed camera to see when light that has passed through an object arrives, and from this information, their algorithm can reconstruct an image of that object. The technique could be useful for improving medical imaging technology, as visible light can carry much larger amounts of information that ultrasound waves of X-rays, as well as for autonomous vehicles, as it could help the car’s computer vision algorithms identify objects through fog or rain.
The Pinal County Public Health Services District in Arizona has partnered with software firm SAS to develop analytics software to help public health officials better monitor and respond to heat-related illnesses. The software analyzes historical data from hospitals and state mortality databases about heat-related medical incidents, such as heat stroke, hyperthermia, and heat shock, to identify when and where heat-related illnesses spike. Health workers can also incorporate demographic data into the software, to identify how factors such as economic status or age contribute to a community’s rates of heat-related illness. With this analysis, the Pinal County Public Health Services District can send out targeted alerts, help emergency responders and health providers better respond to heat waves, and develop more effective educational material about heat-related illness.
Researchers at the University of Southampton have successfully demonstrated how memristors—electrical components that regulate the flow of electricity in a circuit and can “remember” different configurations after being turned off and back on—could be used to power an artificial neural network. Because memristors can remember different patterns, they function more like neurons in the brain than traditional electronic components, making them adept at machine learning applications and performing complex processing with very little power consumption, which can be useful for Internet of Things devices.
The White House has launched 29 new tools for its Opportunity Project, an initiative which focuses on increasing economic mobility through the use of open data. The new tools use government open data from the federal and local governments to address issues limiting economic opportunity, such as making transit more accessible to low-income neighborhoods and helping families better understand information about school quality. Additionally, the White House has announced the Department of Commerce will lead the Opportunity Project going forward, as well as announced a series of new commitments from the public and private sectors to support the project.