This week’s list of data news highlights covers August 5 – 11, 2017, and includes articles about an AI system that detects anthrax and a machine learning algorithm that could lead to the creation of stronger metals.
The U.S. Department of Justice has announced a new initiative called the Opioid Fraud Abuse Detection Unit which will use data analytics to catch bad actors overprescribing or illegally distributing opioids. The unit will analyze data such as physician’s opioid prescription rates, the average age of patients receiving opioids, and regional opioid abuse rates to identify doctors, pharmacists, and others that are contributing to the opioid crisis. The initiative will initially focus on 12 districts in states with high rates of opioid abuse, such as Kentucky, Ohio, and West Virginia.
Researchers at the Korea Advanced Institute of Science and Technology have developed an AI system capable of analyzing bacterial spores and quickly determining the presence of anthrax with 95 percent accuracy. Anthrax contamination has a death rate of 80 percent, making it crucial to quickly detect its presence. Traditional methods for anthrax detection, such as gene sequencing or developing a bacterial culture, can take between several hours and a day, while this AI system can identify anthrax in just a few seconds.
Researchers at the Massachusetts Institute of Technology have developed a device that emits low-power radio frequency signals in a user’s bedroom to detect changes in their pulse and respiration rate as they sleep. The device uses an artificial neural network to analyze changes in how these signals reverberate around the room, not unlike radar, and can identify patterns that indicate changes in a user’s vital signs. In a test, the device was 80 percent accurate, which is comparable to traditional methods that typically need to be administered by a sleep specialist and involve an invasive suite of sensors.
The Atlantic City Police Department is seeing results from a predictive policing technique developed by researchers at the Rutgers School of Criminal Justice, called risk-terrain modeling (RTM), after using it for less than a year. RTM analyzes crime data and factors that make specific places hotspots for crime and prompts police to intervene to make these areas less attractive for criminals, such as by helping local business owners install security cameras. In the first six months of 2016, RTM helped reduce violent crime by 20 percent, while also reducing arrests by 17 percent.
Ireland’s Health Service (HSE) has launched an open data portal called eHealth Ireland containing 300 datasets about Ireland’s health sector, including information about hospital wait times and healthcare provider spending. HSE launched eHealth Ireland to promote transparency, identify inefficiencies, and spur innovation in the sector.
Baltimore, Maryland and Lafayette, Louisiana are deploying networks sensors to study neighborhood-level changes in air quality, including ozone, temperature, humidity, and particulate levels. Baltimore and Lafayette are the winners of the Environmental Protection Agency’s 2016 Smart City Air Challenge, which awarded funding to cities that proposed plans to develop best practices about using sensors to understand air in their communities.
DeepMind has developed an AI system capable of learning to associate audio and visual concepts after watching short segments of video without human instruction. The system first uses two neural networks—one specialized in image recognition and the other in audio recognition—to analyze short video clips. Then, a third neural network compares the audio and video of the clips to identify which images correspond to which sounds, teaching itself to associate these concepts without outside input.
Brigham Young University researchers have developed a machine learning algorithm that can identify relationships between the atomic structure of metals and grain boundaries—physical properties of metals that determine their strength, conductivity, and corrosion resistance. Understanding a metal’s grain boundaries is important for understanding its utility, yet there are no methods for predicting how different atomic structures will influence grain boundaries’ properties. The researchers first built a repository of atomic structures of a wide variety of different metals and alloys and then used their algorithm to identify which physical structures were associated with certain grain boundary properties, which could eventually lead to the development of stronger, lightweight metals that do not corrode.
Scientists at Ben Gurion University in Israel have developed a 3D scanning technique called “dip transform” that uses a pool of water to generate 3D scans of complex objects without the need for expensive optical sensors. The technique uses a robotic arm to slowly dip an object into a pool of water, and a sensor regularly measures how much water is displaced. By dipping the object hundreds of times at different angles, an algorithm can match the displacement measurements on each dip and create a high-fidelity 3D model of the object. This approach can also scan complex shapes that have obscured portions that traditional optical sensors would not be able to see.
Computer scientists at the Costa Rica Institute of Technology have developed an AI system capable of identifying over 1,000 species of plant with close to 80 percent accuracy, outperforming human botanists. Herbaria—scientific collections of plant specimens—often have large quantities of samples, but sorting through all of this information to compare species or identify new plans can be very labor intensive. This AI system could eventually help botanists automate plant identification to aid their research.
Image: Britt Reints.