This week’s list of data news highlights covers August 22 – August 28, 2020, and includes articles about discovering new planets with machine learning and experiencing the Tour de France through augmented reality.
Researchers at the RIKEN Center for Computational Science in Japan have tested the effectiveness of different face mask materials against COVID-19 respiratory droplets using Fugaku, the world’s fastest supercomputer. Fugaku can perform more than 415 quadrillion computations per second, and in numerous simulations, identified non-woven masks as more effective than cotton and polyester masks. Non-woven masks blocked nearly 100 percent of all cough droplets, while cotton and polyester masks only blocked 80 percent.
A team of researchers at the University of California, Davis are using genomic analysis to identify endangered species at risk of contracting COVID-19. The team compared 25 amino acids found in the human ACE2 protein, a protein the virus uses to gain bodily entry, to amino acids found in animal ACE2 proteins. The more amino acids matched, the more likely an animal species is to contract the virus. Preliminary findings showed that 40 percent of animals at risk were threatened species, including chimpanzees, white-tailed deers, and bottlenose dolphins.
The University of California, San Francisco is launching an initiative that offers free genome sequencing to all patients. Participants of the study will receive information on their genetic ancestry and genetic predisposition to cancer, cardiovascular diseases, and neurological diseases. The goal of the study is to build a de-identified biorepository of health information that can be easily analyzed by AI tools to help clinicians better identify the genetic and environmental causes of disease across diverse patient populations.
Astronomers and computer scientists at the University of Warwick in England have discovered 50 new planets using machine learning. The researchers trained the machine learning algorithm using two samples from NASA’s telescope missions: one of confirmed planets and one of false positive planets. They then applied the algorithm to a sample of unconfirmed planets, which resulted in 50 newly confirmed planets of varying size and length of orbit.
Researchers at the University of Central Florida have developed an AI method that detects traces of fentanyl and also teaches itself to detect new derivatives of the drug. First, researchers used a national database of organic-molecules to identify all the compounds that have fentanyl-specific molecules. Then they trained a machine learning algorithm to identify these molecules based on the frequency of infrared light they absorb. The model has an accuracy rate of 92.5 percent.
This year, Tour de France fans can experience the race virtually with augmented reality (AR). Fans will be able to use an AR app to watch the event from a bird’s-eye view, while also exploring the scenic routes taken by competitors. The organizers of the race will also use machine learning to make predictions about a rider’s performance for each stage in the race.
Researchers and engineers at the University of Michigan-Ann Arbor have created a smart water monitoring system called Open Storm. Open Storm uses Internet-connected sensors to collect data about rainfall, water quality, water flow, and soil moisture from drainage infrastructure. The system transmits the information to researchers at the university’s lab for analysis. This allows researchers to use the system’s algorithms to remotely control draining valves and gates. The University of Texas is now using Open Storm to monitor water levels along roadways and alert motorists in the event flooding occurs.
Researchers at the Karolinska Institute and Karolinska University Hospital in Sweden have confirmed three AI algorithms to be as accurate as trained radiologists in detecting breast cancer from mammograms. The researchers compared commercially available algorithms using 8,805 mammograms of women who had undergone breast cancer screening between 2008 and 2015, 739 of whom were diagnosed with breast cancer. Results showed the algorithms diagnosed the same percentage of women with breast cancer as the average radiologist.
LinkedIn has open-sourced a tool that measures fairness in machine learning workflows. The tool, known as LiFT (LinkedIn Fairness Toolkit), can be used throughout different stages of machine learning training systems and can also be used to analyze bias measurements in existing machine learning models.
Carnegie Mellon University and Allies for Children in Pennsylvania, a local child advocacy organization, are using AI to increase the efficiency of school bus food deliveries for students reliant on meal programs. The machine learning algorithm uses data on students’ home addresses and their enrollment in school meal programs to compute a central delivery stop that maximizes the number of students served in the least amount of time.
Image: Martin Biilmann