This week’s list of data news highlights covers October 5-11, 2019, and includes articles about a system that uses AI to see through walls and a new genomics technique that identifies the causes of rare diseases.
Researchers from Owkin, a start-up based in New York that develops machine learning tools for biological research, has developed an AI system that can accurately predict the survival of mesothelioma patients by analyzing tissue samples. The researchers trained the system on data in the French MESOBANK, a database of tissue samples and cell lines. The AI system was able to identify new biomarkers, including features associated with inflammation and cellular diversity, that were predictive of patient survival rates.
The United States Department of Energy has created a new AI research center called the Center for Artificial Intelligence-Focused Architectures and Algorithms. Researchers from multiple government laboratories and the Georgia Institute of Technology will collaborate at the center, which will focus on areas such as the power grid, cybersecurity, and computational chemistry. The center will also focus on how to optimize hardware and software for one each other given a particular problem.
Researchers from MIT have developed a system that uses AI and radio waves to detect what individuals are doing, even behind walls and in darkness, and recreate 3D stick figures of the scene. The researchers trained the system using video images and radio wave representations of the same images, teaching the system to correlate the visible and radio images in the same scene. The system is roughly as accurate as systems that recognize the actions of humans in visible scenarios.
Researchers from MIT and IBM have developed a technique to train computer vision models on devices that have low processing capacities, such as phones. Most computer vision algorithms will process a video by splitting the video into image frames and tracking how objects in the frames change over time, which requires the algorithm to remember each frame in their correct order, a computationally demanding task. The researchers’ technique has the algorithm extract basic sketches of objects in each frame and overlay them on one another, allowing the algorithm to analyze what happened by looking at how objects shift in space instead of remembering what happened in each frame. The technique trains algorithms three times faster than state-of-the-art approaches.
A group led by an individual from Scripps Research has developed a genomics method that can reveal the causes of rare diseases. The technique uses gene transcription data to detect differences in the activity levels of maternal and paternal alleles, which are copies of every gene that most people inherit from their parents. Alleles with significantly different activity levels can be a sign of a disease. The researchers applied the technique to individuals with rare muscular dystrophy diseases, finding it could successfully detect the disease-linked genes.
Researchers from the University of Grenoble in France and Clinatec, a French biomedical research center, have developed a brain-controlled exoskeleton that helped a paralyzed man to walk. The man controlled the skeleton using two sensors implanted close to his brain, which each had 64 electrodes, and an algorithm translated his brain waves into instructions for movement. The man trained the system to understand what he was thinking by practicing with a virtual avatar.
Researchers from the University of Missouri have shown that sweat sensors can help predict outbursts by autistic children. The researchers had autistic children wear wrist and ankle monitors that tracked their electrodermal activity, which is indicative of rising sweat levels. The monitors recorded spikes in electrodermal activity 60 percent of the time the children had an outburst. The research shows how monitoring autistic children’s physical responses can help parents or caregivers intervene before a situation becomes too stressful for a child.
Researchers from MIT and the University of California at San Diego have developed a machine learning system that identified 800 suspicious networks related to IP hijacking, which is when attackers reroute Internet traffic maliciously. The researchers used data from network operator mailing lists and the global routing table to train the system, and the system learned that the use of IP addresses from multiple countries and the quick disappearance of networks are indicators of malicious activity.
Sports technology company SMT has developed a puck and player tracking system for the National Hockey League to help teams gather data about player speed and performance. The system uses infrared and radio-frequency sensors embedded in the pucks and the collars of players’ jerseys and processing devices in upper tiers of arenas that record the coordinates of each sensor hundreds of times a second. The system then uses AI-enabled software that can calculate statistics, such as a player’s top speed or time of possession.
Takeshi Inomata, an archaeologist from the University of Arizona, has used publicly available LIDAR maps published by Mexico’s National Institute of Statistics and Geography to find 27 previously unknown Mayan ceremonial sites that demonstrate a type of construction archaeologists had never seen before. LIDAR helps archaeologists peer through dense forest canopies to find sites, and Inomata used the LIDAR map to find sites made of rectangular platforms that are low to the ground but extremely large, with some sites being as long as two-thirds of a mile, which makes them difficult to detect when viewing them from the ground.
Image: Keith Allison