This week’s list of data news highlights covers September 30 – October 6, 2017, and includes articles about a robotic emergency responder and an effort to use data analytics to identify who turned in Anne Frank.
Volunteers at Columbia University hosted a mapathon to update maps of Puerto Rico to help emergency responders overcome logistical challenges facing relief efforts for Hurricane Maria. The volunteers analyzed satellite imagery of Puerto Rico and made updates to maps of the region on crowdsourced mapping platform OpenStreetMap to inform relief workers about transportation networks as well as highlight any damaged infrastructure that might be inaccessible. Portions of Puerto Rico lacked mapping data even prior to Hurricane Maria, making relief planning efforts particularly challenging.
Agriculture researchers at Pennsylvania State University have developed an AI system that runs on a smartphone and can identify diseases in the cassava plant with near 100 percent accuracy. The researchers developed the system with Google’s open source machine learning platform TensorFlow and used a machine learning technique called transfer learning to train the system on a relatively small training dataset of just 2,756 images of cassava leaves.
Honda has developed a prototype robot called the E2-DR designed to navigate difficult terrain to aid in disaster relief efforts. The humanoid E2-DR can climb stairs and ladders, step over obstacles, and walk over debris thanks to a variety of sensors that help it map its surroundings, including two rotating laser rangefinders, high-resolution and stereoscopic cameras, as well as 3D sensors and cameras on each of its hands.
Retired Federal Bureau of Investigation (FBI) agent Vince Pankoke has recruited a team of investigators, historians, and data scientists to attempt to uncover who tipped off SS soldiers about the whereabouts of Anne Frank and her family while they were hiding from the Nazis in 1944 in Amsterdam, leading to their capture. The team will compile a database of Nazi records and other documents potentially related to the case, such as names of informants and Gestapo officers living in the city. The team is working with Amsterdam-based analytics firm Xomnia to develop an algorithm that can analyze this database for potential clues.
Researchers at the University of Texas at Austin have developed a machine learning system capable of analyzing magnetic resonance imaging (MRI) data of brains of patients with glioma, the most common type of brain tumor. The system can identify the location of tumors and classify types of tissue, such as “edema” (inflammation) or “tumor core.” In a test, the system was able to identify and classify tumors with 90 percent accuracy in less than four hours, comparable to human experts.
Designers at the Copenhagen Institute of Interaction Design has developed a table called The Classyfier embedded with sensors that can identify beverages people are drinking nearby and play corresponding music. The Classyfer uses audio sensors and machine learning to identify characteristic sounds of different beverages, such as cracking open beer cans or clinking wine glasses, and play music that matches moods typically associated with each kind of drink.
A team of researchers led by the Technical University of Denmark used a supercomputer and an algorithmic process designed to mimic natural selection to generate a structure for an aircraft wing that is as stiff as a traditional aircraft wing but two-to-five percent lighter. The researchers began with the design for a wing optimized to minimize drag and maximize lift and digitized it into 1.1 billion 3D pixels, known as voxels. Then the researchers used a supercomputer to simulate how force would impact every single voxel, and repeated this simulation hundreds of times, adjusting each voxel to improve performance each time. The resulting structure resembles the bones in bird wings, and if used could save a plane up to 200 tons of fuel per year.
IBM and scientists at the University of Notre Dame have developed a deep learning system for forecasting waves capable of running on the Raspberry Pi, a tiny low-power computer. Forecasting waves involves factoring in wind, tides, and changes in ocean depth, and typically requires large amounts of computing power. IBM’s new deep learning system can run these simulations 12,000 percent faster than standard forecasting methods, allowing it to generate near-real time forecasts on low-power hardware.
Members of Congress have introduced the Smart Cities and Communities Act of 2017 to provide $220 million annually to upgrade infrastructure with smart technologies, such as sensor networks and analytics systems. The bill, introduced by Senator Maria Cantwell (D-WA) and Representatives Suzan DelBene (D-WA) and Ben Ray Luján (D-NM), would also direct the federal government to coordinate the development of smart city initiatives and proposes measures to improve cybersecurity for smart city technologies.
Las Vegas startup FaceTec has developed a facial recognition system for smartphones that uses just a traditional camera and software, unlike other approaches which also rely on specialized sensors, which could enable it to run on any kind of smartphone. The system requires users to take a series of selfies at different distances from the phone so it can analyze the subtle distortion caused by distance from the lens, and once it learns a user’s face, he or she only has to take two selfies for the system to recognize them.
Image: U.S. Navy.