This week’s list of data news highlights covers November 7, 2020 – November 13, 2020 and includes articles about protecting the Vatican Library’s online collection and launching the world’s first AI-powered satellite.
Scientists from the Nanyang Technological University of Singapore have developed an AI-powered app that helps consumers decide whether meat is fit for consumption. Inside meat packages, barcodes act as freshness fingerprints by changing color in response to the amount of gas being emitted from meat. Researchers trained neural networks to predict the state of decay in meat based on images of colored barcodes. When tested, the system detected spoiled meats in under 30 seconds with 100 percent accuracy and identified fresher meats with 96 percent accuracy.
The Roman Catholic Church has partnered with Darktrace, an AI cybersecurity company founded by mathematicians from the University of Cambridge, to protect the Vatican Library’s 80,000 digitized manuscripts from being deleted, stolen, or held for ransom through cyberattacks. Darktrace has created an AI system modeled on the human immune system that allows it to detect abnormal activity in the digital library and respond to cyberattacks, which come in at a rate of approximately 100 threats per month and are increasing in frequency.
Fujitsu Laboratories, the research and development division of the Japanese IT equipment and services company, and Tokyo Medical and Dental University have used Fugaku, the world’s fastest supercomputer, to analyze cancer genes in less than a day, instead of months. Researchers applied Fugaku to a dataset of 20,000 genes to analyze the relationship between genes found in epithelial cancer cells, which are cancer cells found on the tissue that forms the outer layer of a body’s surface such as carcinoma, and the development of secondary malignant growths. Based on this analysis, researchers used a machine learning technique to predict how the relationship between genes affects the spread and development of cancer cells.
Researchers from Baylor College of Medicine in Texas are using machine learning to improve predictive mortality risks models for cirrhosis, a late-stage scarring of the liver caused by liver diseases, by accounting for more clinical and psychosocial factors associated with cirrhosis mortality, such as excessive alcohol consumption. Researchers trained a set of machine learning models on patients with cirrhosis from 130 hospitals and developed the models to have varying levels of complexity over a wide-range of variables, such as old age and pre-existing liver conditions. When tested, the model predicted one-year mortality risk in patients with 11 percent more accuracy than traditional models.
Researchers at the Massachusetts Institute of Technology and Harvard University are analyzing Reddit posts to measure the impact of COVID-19 on people’s mental health. Researchers used natural language processing on 800,000 posts to quantify the frequency of words used in association with topics such as anxiety, death, isolation, and substance abuse. By identifying changes in the tone and language of posts, researchers discovered that the early stages of the pandemic affected the mental health of users with attention deficit hyperactivity disorder and eating disorders the most.
Researchers at the Rochester Institute of Technology in New York are using Frontera, the eighth most powerful supercomputer in the world, to better detect the disturbances black holes of disproportionate sizes create when they merge. Previously, researchers could only simulate collisions between pairs of black holes that were more or less of equal size to maintain the accuracy of the simulation. However, Frontera provides researchers with the high-performance processing, communication, and memory capabilities they need to accurately simulate a black hole merging with another black hole that is up to 128 times as large.
Researchers from Mount Sinai Hospital in New York City have developed machine learning algorithms that forecast how likely a COVID-19 patient is to experience critical illness or death, based on characteristics such as past medical history and vital signs. Researchers trained the algorithms on 4,000 electronic health records to predict disease progression for specific patients, such as how likely it is that they will need to be intubated. Results showed that high blood sugar and acute kidney injury were the strongest predictors for critical illness, and older age and blood level imbalance were the strongest predictors for mortality.
The Commonwealth Scientific and Industrial Research Organization (CSIRO), Australia’s national science agency, has partnered with Microsoft and Hobart City Council to reduce plastic waste in Australia. CSIRO and Microsoft are applying machine learning to camera sensors in waste traps, which are used by cities to prevent waste from flowing into storm drains, to detect and classify the waste found. CSIRO is also working with Hobart City Council to develop an autonomous sensor network that can report in real-time the amount of waste found in storm drains.
Researchers from Cardiff University in the United Kingdom have developed a machine learning model to recommend which medicines rheumatologists should prescribe to patients with rheumatoid arthritis. To train the model, researchers used 42 variables, such as demographic data and disease activity, as predictors for a patient’s drug response. Using a decision-tree classification approach, which breaks down data into smaller subsets to arrive at a decision, researchers built predictive models that showed whether a variable would positively or negatively affect a patient’s drug response.
Ubotica Technologies, a Spanish computer vision and AI company, has launched the first AI-powered satellite to monitor polar ice caps and soil moisture using images captured by thermal cameras in space. Unlike traditional satellites that require analysts to manually filter out unsuitable images that have been obstructed by clouds, the AI-powered satellite automatically filters such images out, enabling researchers to reduce the amount of data they use in storing and downloading these images. Scientists estimate that this process will reduce bandwidth by 30 percent.
Image: Dominique Devroye