This week’s list of data news highlights covers July 18-July 24, 2020, and includes articles about using machine learning to predict volcanic eruptions and teaching robots common sense.
1. Predicting Volcanic Eruptions
Researchers from the University of Auckland in New Zealand have created a real-time monitoring system that can predict volcanic eruptions hours in advance. The researchers used machine learning algorithms to detect the patterns of seismic activity predating past volcanic eruptions, and found that there is typically an energy burst hours before eruptions. The system generates an alert when the possibility of an eruption is greater than 8.5 percent.
2. Teaching Robots Where the Kitchen Table Should Be
Researchers from Carnegie Mellon University and Facebook have created a system that helps robots navigate by understanding the probable location of objects. The researchers trained the system to understand the difference between items such as an end table and kitchen table, teaching it to discern the relationships between objects and room layouts. This process helps the robot understand that items such as the refrigerator will be in the kitchen.
3. Creating Three Quantum Computing Centers
The White House has announced an investment of $75 million to create three quantum computing centers. The centers will be at the University of California Berkeley, the University of Illinois, and the University of Colorado. The centers will examine topics such as the development of quantum sensors and develop a curriculum for students.
4. Using AI and Satellite Images to Track Illegal Fishing
Researchers led by an individual from Global Fishing Watch, a nonprofit based in Washington, D.C., have used machine learning to identify more than 800 vessels that were illegally fishing off the coast of North Korea in 2019. The researchers used a neural network to analyze satellite imagery to identify the boats, which they traced back to Chinese ports. The researchers estimate the fishers caught 160 tons of flying squid, which has a rapidly declining population.
5. Helping Algorithms Understand Their Environment
Researchers from DeepMind have designed a new method to train reinforcement learning algorithms to learn faster. Instead of learning through trial and error, the researchers’ method first trains an algorithm to determine what its environment allows it to do and not do. For example, the researchers had an AI agent explore its range of motion next to a wall in a 2D environment. Unlike other agents, the AI agent then avoided performing any movements that the wall would impede.
Researchers from the ETH Zurich, a university in Switzerland, and the University of Maryland have identified 37 recently active volcanoes on Venus using thermal and infrared images. The researchers used thermal data collected by the European Space Agency’s Venus Express orbiter to simulate how the planet’s volcanoes would look. They then analyzed infrared images collected by a NASA probe for terrains with specific characteristics, allowing them to map the volcanoes.
7. Creating the Fastest AI Supercomputer in Academia
Nvidia and the University of Florida are partnering to create the fastest AI-enabled supercomputer in academia. The computer will use over 1,000 graphic processing units, which are circuits that can rapidly perform mathematical operations simultaneously. The University of Florida has also committed to hiring 100 additional individuals in AI and related fields.
8. Creating a Database of Hundreds of Years of Flooding in Europe
Researchers from the Vienna University of Technology in Austria have found that recent decades of European flooding have been among the worst in the last 500 years by creating and analyzing a database of legal records, annals, and historical letters. The database contains information about 103 rivers across Europe, allowing them to classify nearly 10,000 floods as notable, great, or extraordinary. The researchers found that only two periods before 1990 to 2016 could have had more floods.
9. Predicting Polar Bear Population Levels
Researchers led by an individual from the University of Toronto have created a model that predicts most polar bear populations will significantly diminish by 2100. The researchers used data about different sub-populations access to sea ice, which the bears use to hunt, and the bears’ energy needs and fat levels to find that most sub-populations are at risk of significantly decreasing. The model predicts that only populations living in the far north of the Arctic will be alive in 2100 if temperatures rise by 4 degrees Celsius by 2100.
10. Mapping Outdoor Junk Food Ads in Liverpool
Researchers from the University of Liverpool have used AI and street-level images to find that advertisements for unhealthy food were concentrated in poorer areas in Liverpool. The researchers used a GoPro while cycling to collect the data and developed an AI-enabled system that could automatically extract and classify outdoor advertisements. The system analyzed more than 10,000 outdoor advertisements.
Image: NeedPix
