This week’s list of data news highlights covers October 7-13, 2017, and includes articles about an AI system that can see around corners and an algorithm that can identify the greenest city in the world.
Researchers at Google have developed a machine learning system called AutoML capable of generating new, high-performing machine learning algorithms. AutoML has created a machine learning image recognition system that can categorize images based on their contents with 83 percent accuracy. The system has also generated a machine learning system capable of performing the more difficult task of marking the position of multiple objects in an image with 43 percent accuracy, surpassing Google’s human-built system that was only 39 percent accurate.
According to a new report from Deloitte, London’s public transit agency Transport for London (TfL) generated £130 million (US $172.9 million) in economic benefits per year by publishing open data. TfL publishes data about transit timetables, transit disruptions, service states, and other operational data online in open formats, which helps power over 600 different transportation apps. Publishing real-time service status and disruption data online saves commuters between £70 million and £90 million (US $93.1 million – $119.7 million) per year in transportation planning costs, and TfL data generates between £12 million and £15 million (US $16 million – $20 million) in added value for private firms per year.
Fashion technology startup Vue.ai has developed an AI system that can analyze an article of clothing and generate an image of a model of any size and shape wearing that clothing. The system uses a machine learning technique called general adversarial networks to predict how clothing would fit on a person and then criticize this attempt, repeating this process until it produces a satisfactory prediction.
Massachusetts Institute of Technology researchers have developed an AI system called CornerCameras that can detect the speed and trajectory of objects hidden behind corners in real time. When objects move, their subtle reflections create shadows called penumbra, which can be observed from around corners, though they are difficult to detect with the naked eye. Using a smartphone camera, CornerCameras analyzes the changes in these shadows to identify how the object creating them is moving. This approach could help give self-driving cars better situational awareness.
Researchers at the University of California, San Diego have developed a system for interpreting a bird’s neural activity to predict the song it will sing 30 milliseconds before it does so. The researchers fed data from electrodes monitoring the portion of a bird’s brain that controls its singing into an artificial neural network. Birds’ vocalizations are distinct, so as a bird sang, the neural network learned to associate specific patterns of neural activity with different songs and could reproduce the songs based on this activity before the bird was able to.
Researchers at the University of Vermont Complex Systems Center, which conducts sentiment analysis of social media posts, has concluded that people are less happy on Mondays than any other day of the week. The researchers developed an algorithm that samples 50 million Twitter posts every day and scores the average happiness of the tweets based on happiness scores the researchers assigned to over 10,000 different words. Their analysis found that Mondays are consistently the least happy, while Fridays and Saturdays are the happiest.
Researchers at the Chinese Academy of Sciences have developed a test that can measure the intelligence of AI-powered digital assistants such as Apple’s Siri and compare these scores against human-level intelligence. Their test showed that of the assistants they analyzed, including those developed by Google, Baidu, Apple, and Microsoft, Google’s digital assistant was the smartest, though it ranks just below the intelligence of a six-year-old human. The researchers have been conducting these tests since 2014 and noted that the intelligence of AI assistants is improving quickly—for example, Google’s assistant’s intelligence score nearly doubled in just two years.
The University of Michigan has launched a precision medicine initiative to analyze genetic, lifestyle, medical, and sensor data to develop new personalized treatments. The first focus of the initiative is to identify risk factors that could indicate if a patient prescribed opioids will become a chronic user, which could help doctors develop pain treatment plans that reduce the amount of opioids they prescribe. The initiative will also launch projects focusing on cancer, mental health, and metabolic diseases.
Artificial intelligence research lab OpenAI had an AI system compete against a copy of itself in virtual environments to have it develop strategy on its own. In one test, the AI systems had to use humanoid models to sumo wrestle against each other, winning points if they manage to force the other out of the ring. Through trial and error the systems eventually developed complex strategies to try and outwit each other, such as lowering their center of gravity to make their stance more stable and moving out of the way at the last second to trick their opponent into running out of the ring.
Researchers at the Massachusetts Institute of Technology have developed an algorithm that analyze images from Google Street View to determine the percentage of each image that contains trees to develop a score called the Green View Index (GVI). By plotting the GVI of every street in a city, the researchers were able to develop and compare GVI percentages for entire cities. Of the cities they analyzes, Singapore has the highest GVI at 29.3 percent green, while Paris has the lowest GVI at 8.8 percent green.