This week’s list of data news highlights covers June 10-16, 2018, and includes articles about IBM’s new chip that could make AI more efficient and a seizure-detecting wearable that could help people manage their stress levels.
Researchers at the Technical University of Dortmund in Germany have developed a machine learning model to predict the outcome of the World Cup. The researchers developed their model using a wide variety of factors that could influence a game’s outcome, such as official team rankings, the average age of a team’s players, and whether a team has a home advantage. The model predicts that Brazil is the most likely to win, with a 16.6 percent chance, followed by Germany and Spain, with 12.8 and 12.5 percent chances, respectively.
IBM has developed a microchip designed to mimic the structure of synapses in a brain, potentially making artificial neural networks up to 100 times more efficient. Though artificial neural networks are typically modeled on the structure of synapses, they usually only rely on standard hardware. Like a brain, the chip uses two types of microelectronic synapses: long-term synapses for memory and short-term synapses for computation. In a test involving image recognition, IBM’s chip was as accurate as a neural network using traditional hardware while only using 1 percent of the energy.
Researchers at the Massachusetts Institute of Technology have developed an AI system called RF-Pose that uses AI to allow radiofrequency sensors to detect people’s movements through walls. RF-Pose relies on an artificial neural network to analyze radio signals that bounce off people’s bodies as they move and then pass through walls, and generate a stick-figure model representing a person’s movements. RF-Pose could allow doctors and caregivers to remotely monitor patients with Parkinson’s disease or muscular dystrophy and track the progression of their symptoms.
DeepMind has developed an AI system capable of spatial reasoning as if it was imagining what a space would look like based on limited data. The system uses a type of artificial neural network DeepMind researchers call a generative query network to observe a scene from multiple angles and then describe what the scene would look like from a different angle, just like humans build mental images of space.
A startup called Veo Robotics is using connected sensors to make industrial robots less dangerous. Veo installs a system of sensors around an industrial robot that can identify nearby objects and track when people walk by, while software predicts where different objects will move to and controls the robot to avoid coming into contact with them. Veo designed its system to retrofit existing industrial robots, so companies could make their robots safer without needing to buy new robots entirely.
The Department of Energy’s Los Alamos National Laboratory (LANL) and satellite imagery analytics company Descartes Labs have developed a model to successfully forecast outbreaks of dengue fever in Brazil. LANL and Descartes Labs built their model with seven years of health, environmental, social media, economic, and other historical data combined with seven years of satellite imagery, allowing the model to identify how certain factors such as changes in the environment correlated to dengue outbreaks. For example, the model shows that the presence of healthy vegetation, which indicates the presence of standing water and thus a breeding ground for mosquitoes, could provide up to five weeks of advance notice of a dengue outbreak.
Researchers at the University of California, Irvine have developed a deep learning system that can solve a Rubik’s cube without any assistance. The researchers gave the system the rules of how to solve the puzzle but did not provide feedback about whether a particular turn of the cube is a good or bad move. Instead, the researchers developed a machine learning technique called autodidactic iteration which enables the system to work backwards from a finished cube to learn what good and bad moves are.
New York startup Nanit has developed a baby monitor that uses AI to analyze an infant’s sleep habits and provide recommendations for how parents could help their children get better nights’ sleep. Nanit uses a Wi-Fi enabled camera that can stream video of a crib to a parent’s smartphone and automatically log sleep activity, such as how long it takes for an infant to fall asleep and how many times they wake up per night.
Crisis Text Line, a text messaging-based crisis counseling hotline, is using AI to analyze users’ texts for words and emojis that could signal that a person has a high risk of suicide or self-harm. Crisis Text Line developed its algorithm by exposing it to data from 30 million past user and counselor texts so it could learn what words correlate to high levels of risk. This approach allows Crisis Text Line to prioritize high-risk individuals and provide them counseling immediately when the hotline is receiving a high volume of texts.
Connected health device company Empatica, which developed a wristband called the Embrace modeled after a smartwatch that has been approved by the U.S. Food and Drug Administration to detect seizures, is using the device to help people manage stress. The physiological factors the Embrace analyzes to detect oncoming seizures, including changes in skin conductivity, pulse, and body temperature, also change in response to stress. Empatica also uses machine learning to learn a wearer’s behavior over time, which Empatica hopes could eventually allow it to predict the onset of panic attacks and other stress reactions.
Image: Danilo Borges.