This week’s list of data news highlights covers January 18-24, 2020, and includes articles about mapping the brain and creating an AI system that can help humans make decisions.
Researchers from Google and Howard Hughes Medical Institute have created the largest high-resolution map of brain connectivity. The map diagrams 25,000 neurons and 20 million connections in a fruit fly’s brain and includes areas associated with memory and navigation. The researchers used an algorithm that analyzed 50 trillion 3D pixels of the brain to trace the pathways of cells.
Researchers from Facebook have developed a reinforcement learning algorithm that helps a robot navigate new environments without a map. The algorithm uses data from a depth-sensing camera, GPS, and compass to help robots reach the desired location 99.9 percent of the time. To develop the algorithm, the researchers had bots take 2.5 billion steps in a virtual environment and continuously culled the slowest bots.
Researchers from the University of Alabama and the University of Texas at Arlington have used deep learning to assess how much a home’s curb appeal contributes to its value. The researchers scored 400 images of properties on a scale of one to four based on their curb appeal to train their algorithm. The algorithm then analyzed Google Street View photos and sales data from nearly 90,000 properties to determine that homes with excellent curb appeal sold for seven percent more than similar homes with poor curb appeal.
Researchers from IBM have developed a neural network that can find relevant facts to support an argument with 95 percent accuracy. The researchers trained the system on 400 million newspaper and journal articles and used crowd workers to label sentences that provided evidence for claims related to hundreds of topics. For example, given the claim “Blood donation should be mandatory,” the system can provide facts such as “blood donors have 88 percent less risk of suffering from a heart attack and stroke.”
Researchers led by an individual from MIT have developed a smartphone app that can track road quality. The app uses a phone’s accelerometer to detect movements created by the road, such as potholes or the roughness of the surface. Poor road quality increases fuel consumption as drivers have to press harder on the vehicle’s accelerator.
Researchers from the Inception Institute of Artificial Intelligence in the United Arab Emirates, the Beijing Institute of Technology, and Stony Brook University have developed an AI system that can remove blur from images, including for human faces. The researchers trained the system using nearly 11,000 images of both blurry and non-blurry images.
Researchers from MIT and the Qatar Computing Research Institute have developed an AI system that uses satellite imagery to create more accurate digital maps. The system can detect features, such as the number of lanes of occluded parts of roads by analyzing visible pixels of the road. For example, if an overpass with two lanes is covering part of a road that visibly shows four lanes before and after the underpass, the system can accurately predict the road underneath the overpass has four lanes.
Researchers from Carnegie Mellon University have developed a system that uses AI to analyze and promote positive comments made on social media. For example, the researchers trained the system on more than 250,000 YouTube comments to teach it to highlight comments that are sympathetic to the Rohingya community, a Muslim minority group that has been persecuted by the Myanmar government. The system complements other approaches that often focus on blocking negative speech online.
Pearl, an AI startup based in California, has developed a system that uses machine learning to detect dental fraud. The system analyzes dental imagery to discern if the dentist had already submitted a claim for the same image and to determine if the image and treatment described on the claim match. For example, the system can detect if the dentist is filing a claim for a dental implant when the image shows the patient only needed a filling.
Derq, a startup based in Dubai and Detroit, has developed a system that uses sensors on roads and AI to notify drivers of potential hazards. The system can detect pedestrians and cyclists that are hidden from the driver’s view and send alerts to a vehicle’s dashboard if it has a vehicle-to-everything antenna. The firm is testing the technology on roads in Detroit and Vienna.