This week’s list of data news highlights covers September 21-27, 2019, and includes articles about an AI system that can predict El Niño and a charity using AI to tackle homelessness.
1. Achieving Quantum Supremacy
Researchers from Google have demonstrated quantum supremacy, which is when a quantum computer performs a calculation that is impossible for classical computers to perform. The researchers used a 53-qubit quantum computer to verify that a random number generator was truly randomly generating numbers. The quantum computer performed the calculation in less than four minutes—the same process would have taken Summit, the world’s fastest supercomputer, 10,000 years.
Researchers in the UK have developed a neural network that can analyze MRI scans of the heart 13 times faster than humans. The researchers trained the network on the MRI scans of 600 patients, finding that the system could determine how well a heart was functioning in four seconds—compared to 13 minutes for humans. In addition, the researchers found there was little difference in accuracy between the network and humans.
3. Identifying Drivers Who Text
New South Wales, Australia, is implementing a system that uses high definition cameras and AI to automatically identify drivers who are texting. During a six month trial between January and June, the system checked 8.5 million vehicles, finding 100,000 drivers were using their phones illegally. Once the system is deployed, humans will verify images of offending drivers.
4. Using IoT to Track Flooding
The Department of Homeland Security deployed more than 100 networked sensors in North Carolina, Kentucky, Virginia, and Maryland to provide alerts regarding flooding during Hurricane Dorian as part of a pilot program. The sensors helped state and local officials identify emerging threats by measuring and reporting rapidly rising water levels.
Researchers from Chonnam National University in South Korea and Nanjing University of Information Science and Technology in China have developed an AI system that can predict the climate cycle known as El Niño, which occurs every two to seven years and often results in droughts and flooding in different parts of the world, 18 months in advance. The researchers trained the system on simulated data for El Niño between 1871 and 1973 and tested it on real data from 1984 through 2017. The system can predict the event as far out as 18 months with 74 percent accuracy, while previous models could only make predictions one year in advance.
6. Systems Using AI Can Make Medical Diagnoses as Accurately as Doctors
Researchers led by an individual from a semiautonomous unit within the UK’s National Health Service have found that AI systems can be as effective as medical professionals at diagnosing diseases. The researchers analyzed 14 studies, finding that AI systems correctly identified the disease 87 percent of the time, compared to 86 percent for medical professionals. In addition, the systems correctly identified patients as disease-free 93 percent of the time, compared to 91 percent for medical professionals.
7. Verifying the Safety of Medical devices
Researchers from Stanford University have developed an AI system that can analyze electronic health records to assess the safety of medical implants. The system links implanted devices with patient infection rates, patient pain levels, and length of use before a replacement is necessary. The researchers tested the system on the EHR data of hip replacement patients, finding the system could when identify complications from a device occurred with 96 percent accuracy.
8. Simulating Basketball Plays
Researchers from National Chiao Tung University in Taiwan and NVIDIA have developed an AI system that allows users to design basketball plays and simulate how real offensive and defensive players would move during the play. The researchers trained the system’s generative adversarial network on NBA player movement data from the 2015-16 season. The system could help coaches understand how defensive players would react to their plays.
A researcher based in Madrid has developed Guard, a tool that uses AI to rate an application’s privacy policy. Guard’s users train the tool by being shown snippets from different policies and choosing their preferred choice, thereby teaching the tool the types of policies humans find the most acceptable. Guard has already provided ratings for apps such as Waze, LinkedIn, and Telegram, which received an A+.
10. Using AI to Tackle Homelessness
Researchers from the Alan Turing Institute, the UK’s national institute for data science and artificial intelligence, have developed a machine learning model that can help local organizations prioritize which homeless individuals to help. Individuals in the UK can call StreetLink, a charity that connects homeless individuals with support services, if they see an individual who needs help. However, StreetLink’s small staff struggles to decide which individuals to prioritize and only finds one individual for every seven calls because callers often provide incomplete details about the homeless individual’s health, clothes, location, and gender. The machine learning model uses past decision data to automatically categorize which alerts StreetLink should prioritize.
Image: Airman Sadie Colbert
