This week’s list of data news highlights covers October 10, 2020 – October 16, 2020 and includes articles about ending world hunger with the assistance of machine learning and measuring how effective planting trees is for combating climate change.
Command Sight, a wearable technologies company, is developing augmented reality (AR) goggles for military dogs, enabling soldiers to guide them remotely during dangerous operations. Currently, soldiers remotely communicate with military dogs using cameras and walkie talkies, however the commands can be confusing and misleading for the animals without accurate visual cues. The new system uses AR to provide a visual cue for the canines while a camera in the goggles offers a live feed of the dog’s point-of-view for the handler to monitor, enabling soldiers to use the surrounding environment to provide more contextual commands.
Scientists at the University of Hawaii’s Manoa Institute for Astronomy have produced a 3-D map of celestial objects in the universe using neural networks, an algorithm that mimics the way neurons work in the human brain. To train the model to predict the distance of celestial bodies and identify what they are, the team fed the AI system four million spectroscopic measurements, which measure distance based on the interaction between light and matter. The AI system accurately classified galaxies and stars 98 percent of the time and quasars, which are large celestial objects found at the center of galaxies that emit large amounts of energy, 97 percent of the time.
Researchers at the Chan Zuckerberg Biohub are using machine learning to estimate the number of unreported cases of COVID-19 from genomic mutations. Their theory is, because the virus can mutate when it spreads through a population, understanding how many times the virus has mutated between two observed genomes can indicate how many missing transmission links there were in-between them. To test how different levels of unreported missing transmission links might explain the levels of transmission we see in the real world, the researchers simulated different models of how the disease might spread, based on different levels of mutation speed. They then used machine learning to explore different explanations of the data. Researchers estimated that in the first few weeks of the outbreak, 98 percent of infections were undetected in 12 locations in Europe, China, and the United States.
Researchers at Riken’s Center for Computational Science and Kobe University in Japan have used Fugaku, the world’s fastest supercomputer, to show the effects of humidity on the spread of COVID-19 particles. The researchers conducted simulations of virus-like particles flowing from infected people in different indoor environments, and discovered that in environments with humidity levels of less than 30 percent, more than double the amount of airborne particles were transmitted compared to environments with humidity levels of 60 percent or higher. This indicates risks of coronavirus contagion are higher in dry, indoor conditions during the winter months.
Facebook and Carnegie Mellon University are using AI to find new electrocatalysts, which are materials that can convert excess solar and wind power into more easily stored fuels, like hydrogen and ethanol. Existing electrocatalysts, such as platinum, are rare and expensive, and finding new ones is difficult. To discover new low-cost catalysts from existing metals, researchers must test the viability of thousands of combinations of elements, but each calculation can take hours or even days. Researchers are using machine learning to better explore the full field of possible catalysts by teaching AI models to approximate the energy and forces of molecules based on past data.
Researchers from Ceres2030, a group of climate, social, and agricultural scientists, have used machine learning to analyze 500,000 scholarly articles in order to better understand agricultural practices and development interventions to address world hunger in this decade. The machine learning model scanned each article for keywords that indicated connective themes among different targeted policy interventions. The AI analysis showed that the global economy needs to allocate $14 billion annually in order to end world hunger by 2030, double the amount allocated now. Additionally, since 75 percent of small farmers are located in water-scarce areas, the analysis also pointed to increasing investments in data networks so that farmers can apply fertilizer between rains to minimize run-off.
An international team of scientists have used a supercomputer to measure how effective afforestation, the process of planting trees in areas that were previously unforested, is in combating climate change in Europe. The scientists simulated nine different regional climate models with different levels of afforestation to see how changes in the land affect how much solar radiation is reflected back to the Earth’s atmosphere. Simulations showed that across all nine models, Northern European regions with maximum forestation were 0.2 to 1 degrees Celsius warmer in the winter and spring compared to grassland areas.
Scientists from Oxford University have used machine learning to discover that lions have uniquely identifiable and trackable roars. The team trained a pattern recognition algorithm to recognize lions’ roars using audio data and GPS movement, and discovered that every lion’s roar produces a unique frequency shape. When the team tested the algorithm on new recordings, it accurately matched roars to individual lions with 92 percent accuracy. This tool can help wildlife conservationists track the population of the 20,000 vulnerable lions remaining in the wild to keep a record of poaching activities.
Microsoft has developed an AI image-captioning system that can generate alt text, a written description for images on web pages and documents, for people with low vision or blindness more accurately than previous models. Software engineers first trained the system using a dataset of images and corresponding phrases and then fine-tuned it using a dataset of images that were fully captioned, enabling the system to compose full sentences for alt text. The system is now integrated into SeeingAI, an app that uses a smartphone camera to read text, identify people, and describe objects and the environment, and will be added to Microsoft Word, Outlook, and Powerpoint later this year.
Scientists from the University of Melbourne in Australia and the University of Otago in New Zealand have developed an AI system that uses images of retinal blood vessels to predict a person’s risk of cardiovascular disease. To automate the technique, scientists trained the AI system using 70,000 retinal photographs from 15 multi-ethnic and multi-country databases of people. Results showed that the AI performed the same or better at predicting a patient’s risk of the disease by using retinal images compared to doctors that used traditional testing methods to diagnose, such as measurements of blood pressure, body-mass index, cholesterol levels, and blood sugar levels.
Image: Wade Lambert