This week’s list of data news highlights covers March 2-March 8, 2019, and includes articles about an AI chatbot matching patients to clinical trials and using supercomputers to find the best materials for solar cells.
The University of Pittsburgh Medical Center developed an algorithm with machine learning that decreased its patient rehospitalization rate by about 50 percent. The algorithm analyzes electronic health record data to identify patients at high risk of being hospitalized within seven and 30 days of discharge, allowing clinicians to adjust the care of high-risk patients to reduce the likelihood such patients return to the hospital.
Researchers at Ohio State University, security firm FireEye, and engineering and IT company Leidos have developed an AI system that can analyze millions of tweets to identify security flaws. The researchers trained the system on 6,000 tweets that discussed security vulnerabilities, enabling it to identify most security flaws days before the U.S. government lists them in the National Vulnerability Database. In addition, the AI system uses natural language processing to predict the severity of the vulnerabilities with 78 percent accuracy.
IBM has created a new way to measure the progress of quantum computing called “quantum volume.” Instead of solely measuring the number of qubits, which may not accurately reflect a machine’s performance, quantum volume counts qubits as well as factors such as error rates and the quality of connectivity between qubits.
Microsoft has developed an AI chatbot called Clinical Trials Bot that makes it easier to match patients to clinical trials. With roughly 50,000 clinical trials occurring worldwide, it is difficult for both doctors and patients to find trials that have criteria a patient meets. Clinical Trials Bot simplifies this process by allowing individuals to perform a search such as “trials for a 52-year old California female with breast cancer,” and responds with questions to refine the list of potential trials.
Japanese startup Vaak has developed AI software that detects shoplifters by analyzing security camera footage for signs of potentially suspicious movements, including fidgeting and restlessness. The software alerts store staff via a smartphone app when it detects a potential shoplifter, allowing the employee to ask if the customer needs help, which can reduce the chance a theft ever occurs.
Researchers from the U.S. Department of Energy’s Argonne National Laboratory and the University of Cambridge used machine learning and a supercomputer to identify promising materials for dye-sensitized solar cells, which convert sunlight to electricity. Different dyes absorb sunlight at different rates, but it can be challenging to identify the most effective dyes due to the high number of potential materials that can create a dye. The researchers used Argonne National Lab’s Theta supercomputer to identify five dye materials that provided the best combination of performance, cost, and low environmental impact from a possible 10,000 candidates.
Researchers from Google and ETH Zurich, a Swiss university, have developed an AI model that generates realistic-looking images after training on data with significantly fewer human-created labels than other models. The researchers trained their model by using self-supervised learning, which inferred labels on unlabeled images to train a generative adversarial network. The researchers’ model was able to achieve a high level of performance using only 20 percent of the labels in Imagenet, a large image dataset.
Beth Israel Deaconess Medical Center, a hospital affiliated with Harvard University, is using AI from Amazon to boost its operating room capacity and make tasks like patient scheduling more cost-effective. Amazon’s AI software predicts when patients are likely to miss appointments, alerts staff to missing paperwork, and helps physicians book and estimate operating room times more precisely. Engineers from the hospital and Amazon analyzed data about surgeries to create the hospital’s new scheduling system, which has increased the hospital’s operating room capacity by 30 percent.
Rescue crews used drones equipped with thermal-imaging technology to search for individuals in collapsed buildings in the aftermath of a tornado that killed 23 people in Alabama. While it is unclear if the drones helped find anyone, the drones did help rescuers confirm that they had not missed anyone during their search.
A new startup called Kobold is using AI to analyze geological, physical, and chemical datasets to find potential cobalt mining sites outside of the Democratic Republic of Congo (DRC). Most lithium-ion batteries, which power devices such as smartphones, use cobalt, and the DRC provided 68 percent of the world’s mined cobalt in 2017. It is common, however, for children in the DRC to mine cobalt in toxic environments.