This week’s list of data news highlights covers September 19, 2020 – September 25, 2020 and includes articles about detecting signs of arthritis early with machine learning and using AI to win an esports tournament.
Researchers at Riken’s Center for Computational Science in Japan have measured how effective plastic face shields are at preventing the spread of coronavirus droplets using Fugaku, the world’s fastest supercomputer. After running simulations, researchers discovered that nearly all airborne droplets smaller than 5 micrometers, which is five millionths of a meter, escaped through the shield. Although plastic shields are ineffective for trapping small droplets, the researchers found the shields were effective for trapping airborne droplets larger than 50 micrometers.
A multi-disciplinary team of European researchers are developing an AI tool that detects implicit forms of anti-Semitism in English, French, and German. Using texts and images found online, the tool distinguishes words, patterns of words, and memes for anti-Semitism based on the use of coded language and hidden phrases. Once the tool is developed, the team foresees social media platforms implementing it.
Researchers at the University of Pittsburgh School of Medicine and Carnegie Mellon University in Pennsylvania have created a machine learning algorithm that detects signs of osteoarthritis, a type of arthritis where joint cartilage becomes worn down, in presymptomatic patients. Currently, doctors detect osteoarthritis when irreversible bone damage occurs. To enable early detection before damage happens, researchers trained an algorithm to recognize cartilage patterns of wear, using MRI scans from the knees of osteoarthritis patients. When tested on 86 patients with no symptoms or visible signs of cartilage damage, the model predicted osteoarthritis with 78 percent accuracy 3 years before patients experienced symptoms.
Researchers at the Stevens Institute of Technology, a research university in New Jersey, have developed an AI tool that diagnoses Alzheimer’s disease using sentences written by those with Alzheimer’s and those without. The algorithm assigns each sentence a numerical value, and uses the value to analyze complex sentences. This process more efficiently identifies sentence structures and linguistic patterns characteristic of subjects exhibiting signs of Alzheimer’s. Because the AI tool can explain its conclusions, experts are able to check the accuracy of the diagnosis by approximately 95 percent.
Kroger, a U.S. supermarket company, has partnered with Everseen, an Ireland-based AI software company, to streamline self check-out processes at grocery stores. Everseen’s AI tool uses computer vision to watch self-checkout video recordings in real time. When an error occurs during self-checkout, such as an item failing to scan, the tool alerts store associates via a mobile device so that they can intervene. Using the tool has allowed Kroger store employees to better identify customer issues and more quickly respond to instances of theft, thereby enhancing customer experiences and reducing product losses.
Researchers from the University of California San Diego School of Medicine are using natural language processing, a field of AI focused on analyzing natural language, such as speech and text, to predict loneliness in older adults. The researchers recruited 80 participants to participate in individual interviews. After analyzing a transcript of the interviews, the AI tool accurately predicted a person’s loneliness with 94 percent accuracy compared to a standard loneliness assessment. The tool further revealed differences in the way men and women talked about loneliness, such as women being more likely to directly acknowledge feeling lonely and men being more likely to use fearful and joyful words to express loneliness.
Hexagon, a Swedish technology company, has created an AI tool to assist emergency call operators by identifying patterns, similarities, and anomalies in incident reports. The tool looks for keywords, recurring locations, statistical abnormalities, and weather patterns, and alerts relevant operators about trends in calls that point to major incidents or linked incidents, like a fire or multi-bank robbery. Depending on the magnitude of the incident, operators can share this alert with first responders to coordinate the most efficient response.
Researchers from Monash University, Alfred Health, and the Royal Melbourne Hospital in Australia are using machine learning to detect epilepsy, a brain disorder that causes seizures, from recordings of electrical brain activity. The researchers applied over 400 recordings of patients with and without epilepsy to the machine learning model to identify abnormal spikes in electrical activity among neurons, a sign of epilepsy. With the model, doctors can more quickly diagnose patients with epilepsy.
The Defense Innovation Unit, an organization of the U.S. Department of Defense, is using AI to detect coronavirus early in asymptomatic military personnel with biometric data from wearable devices, like smartwatches. Scientists used biomarkers like pulse and blood pressure from 41,000 COVID-19 cases to train the AI tool to identify physiological responses specific to infectious virus agents before coronavirus symptoms emerge. Since being piloted in select military bases, the system has caught multiple cases of COVID-19 early.
Vietnam is hosting the first Reinforcement Learning Competition, an esports tournament where competitive video game teams must use AI in order to defeat the opposing team. Each team must create an autonomous player based on a reinforcement learning algorithm, a machine learning technique that uses trial and error in order to improve its decision making abilities. After each tournament round, advancing teams receive consultation from AI experts about how to best update or change their programming code so that the autonomous player can continue winning.
Image: Jan Vašek