This week’s list of data news highlights covers April 11-17, 2020, and includes articles about remotely monitoring COVID-19 patients and using AI to evaluate NFL prospects.
1. Remotely Monitoring a COVID-19 Patient
Doctors in Boston used an AI-enabled tool created by MIT researchers to remotely monitor a COVID-19 patient’s health. The tool, which resembles a box and does not touch the patient, emits wireless signals that reflect off the patient. An algorithm analyzes the reflected signals to determine an individual’s heart rate, movement, and sleep quality.
2. Transforming Objects in Images
Researchers from Adobe, Nvidia, and Aalto University in Finland have created software that enables users to easily alter the shape of objects in images and change lighting conditions. The software uses generative adversarial networks (GANs) to make the changes, which can generate realistic images. The researchers discovered which layers in GANs control large-scale variations in the images, allowing them to controls the alterations GANs create.
Researchers from MIT are using supercomputing resources from Amazon Web Services to develop a drug that treats COVID-19. Coronavirus causes COVID-19 by binding to ACE2 receptors in the body. As such, the researchers created a machine learning model, which runs on Amazon’s remote-access supercomputing resources, that simulates how coronavirus might mistakenly bind to a decoy version of the protein. The researchers plan to test the drug in mice by June.
4. Simulating Bad Actors on Facebook
Facebook is using bots it trains using reinforcement learning to determine which potential features on the platform could make it more difficult for bad actors to violate its community guidelines. The bots optimize their behavior for a particular objective, such as portraying a hacker trying to access an individual’s private photos. Facebook’s system automatically adjusts parameters, such as an account’s privacy settings, to determine which settings can best block or constrain a bad actor.
5. Discovering Why Some People Develop Severe COVID-19 Symptoms
Medical testing company LabCorp and Ciox Health, a company that helps firms manage health data, have partnered to determine why COVID-19 affects patients differently. The firms are pooling more than 650,000 COVID-19 tests and relevant medical records. Ciox Health will use natural language processing to extract information from doctor’s notes in the records, and the companies will then train deep learning models on the data.
Researchers from Intel and Delft University in the Netherlands have shown that qubits can operate on a chip at temperatures higher than -273 degrees Celsius. The need for quantum systems to function at extremely low temperatures forces quantum physicists to store qubits in powerful refrigerators and connect each qubit via a wire to controllers that sit outside the refrigerator. This approach is tedious, but the ability of qubits to function at higher temperatures can allow for quantum hardware and circuitry to sit on the same integrated circuit, making it easier to create quantum computers with more qubits.
7. Applying Evolution to Algorithm Development
Researchers from Google have developed an AI system that can create algorithms to solve problems, such as recognizing an object in an image, with little input. The system approximates evolution by randomly combining mathematical operations into algorithms then iteratively deleting and altering them until it finds the best solution.
Microsoft has created an AI feature for its Teams collaboration tool that suppresses unwanted and unexpected background noises in video calls, such as dogs barking. The firm trained the feature’s AI model on datasets of individuals talking with and without unwanted sounds, including data from audiobooks and YouTube.
9. Using AI to Analyze NFL Prospects
Slants, a sports analytics startup, has developed software that can analyze footage of football players to provide teams relevant metrics, such as a player’s speed or how much a wide receiver has separated from a defender. The software uses computer vision to help National Football League (NFL) teams better analyze college players’ performance and athleticism as, unlike NFL players, their uniforms don’t include motion-tracking chips. The data could also replace current NFL practices for scouting college players, such as tracking a player’s speed using handheld stopwatches.
10. Using 911 Calls to Map COVID-19 Clusters
Corti, a startup based in Copenhagen, has developed software that uses AI to analyze interviews in text, audio, or video formats, enabling emergency call centers to detect if a patient has a high-risk of having COVID-19. The Seattle Fire Department is helping the city map suspected infection hotspots by using Corti’s tool to analyze data on 911 callers’ symptoms. The tool has alerted the city to elderly care facilities with numerous cases.
Image: Tammy Anthony Baker