This week’s list of data news highlights covers January 30, 2021 – February 5, 2021 and includes articles about using AI to identify dark matter and predicting how policy changes can impact COVID-19 vaccine distribution.
Researchers at IBM and the University of Bern in Switzerland have applied transformers, a machine learning algorithm used for interpreting sequences of information, to classify chemical reactions. The algorithm generates unique numerical codes that categorize reactions by type and also identifies broader categories, such as carbon bond formations, that scientists can use to search for similar reactions in databases. When tested, the algorithm classified reactions with 98 percent accuracy.
Researchers from the University of San Francisco have developed neural networks that can identify gravitational lenses, which are celestial objects that bend the direction of light traveling from more distant galaxies. Gravitational lenses are important because the way they bend light can help researchers identify dark matter. The team trained neural networks to detect gravitational lenses from 21,632 images, 632 of which were gravitational lenses. When they applied the neural networks to a giant map of the sky, the networks detected 1,210 potential gravitational lenses, more than double the number known today.
Google has embedded an AI tool that can measure vital signs within its newest Pixel phone model. When a user places their finger on top of the rear-facing camera and positions their head and torso towards the front-facing camera, the tool uses camera sensors to detect their pulse and computer vision technology to analyze changes in their coloring. The tool also uses computer vision technology to measure a user’s breathing rate by detecting small movements in their chest. When tested, the tool measured respiratory rates that were accurate within one breath per minute.
Researchers at IBM have developed an AI algorithm that can predict the onset of Alzheimer’s disease years before patients exhibit symptoms. The researchers trained an algorithm to identify signs of Alzheimer’s, such as increasingly repetitive language, typos, or missing words, from the writing samples of 80 study participants, half of whom had Alzheimer’s and half who did not. When tested, the algorithm predicted which participants would develop Alzheimer’s with 75 percent accuracy.
Google has developed an information portal, vaccine scheduler, and predictive analytical model designed to relieve the workload of government officials managing vaccine distribution logistics. Currently, North Carolina has integrated the information portal into their public health system, which answers frequently-asked questions about COVID-19 and offers a self-assessment for people who are unsure if they are eligible for a vaccine in their region. If found eligible, users are directed to call an AI-based virtual agent to schedule a vaccination appointment.
Researchers from the University of British Columbia in Canada have developed an AI system that can identify misinformation. To train the system, the researchers created a training dataset by swapping out nouns, verbs, and adjectives from Arabic news articles with similar words that varied in their manipulation. The researchers then tested the system on 4,500 Arabic articles from a previous study, 1,475 which were fake, and found that the system detected fake articles more frequently than humans.
Researchers from the University of Alberta in Canada have developed a machine learning algorithm that can identify healthy first-degree relatives of schizophrenia patients who are most vulnerable to developing the condition. During their lifetime, first-degree relatives of schizophrenia patients have a 19 percent risk of developing schizophrenia compared to the 1 percent risk of the general population. The tool analyzes features on magnetic resonance images of patient’s healthy relatives to identify markers for schizophrenia.
Getafe CF, a soccer club in Spain, has partnered with Zone7, a Californian company that uses AI to enhance athlete performance, to predict players who are at risk of injury. Getafe CF sends data from soccer training sessions and matches to Zone7, which then uses an algorithm to analyze the information and sends back daily emails about players who may be at risk for injuries, such as spraining a muscle or tearing a ligament. Since the partnership, Getafe has experienced a 40 percent reduction in the number of injuries.
Clinicians at the University of Michigan have developed a tool called CASSY that predicts a teen’s suicide risk. Unlike existing tools, CASSY asks adolescents questions about a range of factors that may put them at risk, such as sleep disturbance, trouble concentrating, and depression. Based on their answers, CASSY calculates a score for their suicidal risk and alerts providers of which patients are in need of follow-up interventions. When tested on responses from 2,754 teens, 165 of which attempted suicide over a three-month period, CASSY predicted the risk of suicide attempts with 88 percent accuracy.
Researchers from Queen Mary University in London have developed an AI system that uses radio wave signals and neural networks to identify the heart rate and breathing pattern associated with anger, sadness, joy, and pleasure. To build the system, researchers bounced radio waves off study participants as they watched videos and fed the returning signals into a neural network. The signals vary depending on an individual’s internal heartbeat movements, inhalations, and exhalations. The neural network then processes and analyzes these signals to reveal hidden data patterns associated with specific emotional states.
Image credit: Stock Snap