This week’s list of data news highlights covers January 6 – 12, 2018, and includes articles about a database of rhino DNA that can help fight illegal poaching and an autonomous weed-killing robot.
1. Measuring the Philippine’s Economy In New Ways
The Philippine Central Bank (BSP) is analyzing data from search engines, job boards, and proprietary sources to supplement official data and get a more timely and accurate measure of the economy. BSP is also developing a sentiment analysis tool that can evaluate how news articles describe the economy based on the use of certain keywords. BSP is the latest central bank in Southeast Asia to turn to alternative data sources to measure the economy, following in the footsteps of Indonesia and Thailand which have both launched similar programs.
2. Improving Transportation Efficiency with Supercomputers
The U.S. Department of Energy’s Vehicle Technologies Office has launched two new initiatives to use supercomputers at the National Laboratories to support research that could increase the energy efficiency of transportation. The first initiative, called “Big Data Solutions for Mobility,” will develop algorithms and data science tools that can help researchers better model and analyze real-time transportation data. The second initiative, called “High Performance Computing for Mobility,” will develop partnerships with transportation authorities, companies, and others to provide them with access to supercomputing facilities and support projects to identify opportunities to improve transportation energy efficiency, such as by optimizing traffic control systems with AI.
3. Using a Robot to Keep Shelves Stocked
Robotics company Bossa Nova has developed a robot that can autonomously roam grocery store aisles and continuously take 2D and 3D images of items on shelves to keep tabs on when different products are running out of stock. The six-foot-tall robot uses LIDAR to map a store and avoid obstacles, such as shoppers, and it can detect when items are ordered incorrectly in addition to monitoring how many of each item are left on shelves.
4. Fighting Poaching with a Rhinoceros DNA Database
Researchers at the University of Pretoria in South Africa have concluded that a database called the Rhinoceros DNA Index System (RhODIS), which contains genetic samples from thousands of live rhinos and stockpiles horns, is a reliable forensic resource for prosecuting poachers illegally killing the animals. 120 criminal investigations have used samples from RhODIS as evidence since its launch in 2010. By analyzing hundreds of samples from RhODIS, the researchers found that the data could be used to correctly identify whether a horn was from a particular rhino with an error rate of less than one in several million, proving RhODIS’ validity as a forensic tool.
5. Putting Robots to Work on the Farm
Agriculture robotics company Blue River Technology is developing a robotics system called “See & Spray” that uses AI to differentiate between crops and weeds and spray small amounts of herbicide directly on weeds, which could substantially reduce the amount of herbicide needed to grow crops. For example, one acre of cotton can require 20 gallons of herbicide, but See & Spray can achieve the same effect with just 2 gallons.
Researchers at Uber have developed a new method for neuroevolution, which involves mutating an AI system’s neural networks to select the ones best suited to solve a problem. Neuroevolution is not a new technique but it typically is only useful for simple neural networks trying to solve easy problems. The researchers studied a variety of different ways to improve upon neuroevolution, and they were able to outperform AI systems that used traditional training methods in some Atari games. In addition, they could control a robot using a neural network developed through neuroevolution that was 100 times larger than any other neural network developed that way.
The University of Iowa’s Iowa Flood Center have deployed 250 water-level sensors on bridges throughout the state and are developing a system called Flood AI to monitor these sensors and make data about flood risks easily available to the public. The sensors send data about water levels every 15 minutes to the Iowa Flood Center and state and federal agencies, which use it to forecast flood risks. Flood AI will allow members of the public to find out about water levels and flood risk through intelligent assistants such as Amazon’s Alexa.
Researchers at Kyoto University in Japan have developed an AI system that can analyze magnetic resonance imaging (MRI) scans of a person looking at an image and recreate an approximation of that image with a high amount of detail. AI systems have previously been able to translate MRI scans into visualizations of what a person sees, but only for simple binary images such as geographic shapes. The researchers’ new system can reproduce a much greater amount of detail to create “hierarchical” images, which contain complicated structures and multiple layers of color.
9. Investigating Underwater Volcanoes With an Autonomous Submarine
Volcanologists at the University of Tasmania deployed an autonomous submersible vehicle called Sentry to scan the seafloor around underwater volcanoes and generate high resolution maps. Underwater volcanoes generate 80 percent of the Earth’s volcanic activity, but they are difficult to study since many are too deep, where water pressure is too high, for traditional survey methods. Sentry’s maps have given researchers unprecedented insights into the structure of underwater volcanoes and how lava flows function underwater.
10. Algorithmically Targeting Cancer Treatments
A group of researchers led by Singapore’s Agency for Science, Technology, and Research have developed a system called ConsensusDriver that can pull information from multiple different recommendation algorithms and generate a consensus recommendation about treatment targets for cancers more accurately than any one algorithm. Sequencing cancer cell’s DNA gives researchers the opportunity to identify potential drug targets, but only certain mutations in a cancer’s DNA drive tumor growth and determining which of these mutations to target can be challenging. In a test on patients who all could be treated with existing drugs, ConsensusDriver identified the correct treatment for 80 percent of patients, while individual algorithms could only identify correct treatments for up to 60 percent of patients.
Image: Don Becker.