This week’s list of data news highlights covers July 29 – August 4, 2017, and includes articles about an effort to use AI to research treatments for aging-related diseases and an analytical model that can identify overpaid soccer players.
Several of Sweden’s largest banks have developed AI-powered virtual customer assistants to operate more efficiently and boost customer satisfaction. The banks, including Nordea Bank, SEB AB, and Swedbank, developed the assistants to serve as chatbots that can access client data and quickly address simple client requests, which allow human employees to spend more of their time on more complicated services. With more user-friendly customer support, the banks could remedy problems that have caused customer satisfaction in Swedish banks to fall to a 20-year low.
The University of Copenhagen’s Center for Healthy Aging has partnered with Baltimore-based medical software company Insilico Medicine to use artificial intelligence to advance research into cures for aging-related diseases such as Alzheimer’s and Parkinson’s. The initiative will focus on using AI to identify molecules that can stimulate the repair of DNA, as many aging-related diseases are linked to genetic factors. Researchers will test molecules identified by the AI system to see if they can contribute to treatments that slow the onset of these conditions.
YouTube has announced that it is expanding its efforts to use machine learning to automatically detect and filter terrorist videos on its site. YouTube users upload 400 hours of video to the site every minute, making it impossible for humans to reliably catch extremist content. YouTube partnered with groups including the Anti-Defamation League and the Institute for Strategic Dialogue to help develop a machine learning system that can automatically flag this content when it arises. In the past month, the system caught and removed 75 percent of the videos containing extremist content before humans could even flag them.
A team of scientists from the Carnegie Institute for Science have developed a method for predicting the locations of undiscovered mineral deposits. Their approach uses an analytical technique called network theory traditionally used to study networks such as a disease’s spread or the Internet. The scientists used millions of data points about where minerals have already been discovered to develop a model that can indicate where other mineral deposits are likely to be. The scientists also used this technique to predict the existence of 145 unknown carbon-containing minerals, 10 of which have since been discovered.
Researchers at Google and the Massachusetts Institute of Technology have developed a machine learning algorithm capable of retouching photos on smartphones before people take them. The researchers trained neural networks on 5,000 different photographs, each touched-up by five different photographers, to teach the algorithm what improvements, such as adjusting brightness or saturation, make a photo look better. While most smartphones and cameras automatically apply adjustments to photos in real-time, they apply the same general improvements to every image, whereas this approach tailors its adjustments to each picture.
Game developer Remedy has partnered with Nvidia to develop a deep learning system that can translate video of actors into 3D animations with realistic facial animations much more quickly than traditional motion-capture techniques. Animating facial expressions is labor-intensive and it can be incredibly difficult to produce realistic-looking animations. The system can generate realistic animations based on 10 minutes of video footage of an actor with much less input from animators, as well as sync an animation’s facial expressions to an audio recording of the actor.
Computer scientists at Lawrence Technological University in Michigan have developed a model that uses machine learning to identify whether a professional soccer player is overpaid or underpaid for their skill level. The model analyzed 55 attributes about 6,082 European soccer players, such as ball control, aggression, and accuracy, as well as their salaries, to develop a representation of what salaries should correspond to different levels of skill. According to the model, Lionel Messi was the most overpaid for the 2016-2017 season, and Bernardo Silva was the most underpaid.
Entertainment software company Gracenote is developing machine learning systems to detect emotional qualities in music. Gracenote developed a taxonomy of over 400 moods, such as “sultry” and “gentle bittersweet,” and manually annotated 40,000 songs in its 100 million song database to train its system. The system analyzes each song by breaking it into 700-millisecond clips to identify specific acoustic qualities and match them to moods it previously associated with those qualities.
Researchers at the Massachusetts Institute of Technology have developed an algorithm named DeepMoji that can identify the emotional subtext in tweets, including sarcasm, better than most humans. The researchers trained DeepMoji on 1.2 billion tweets that already included an indicator about their emotion—emojis—to teach it to associate emotional meaning with text. DeepMoji can detect sarcasm in tweets with 82 percent accuracy, whereas humans could only detect it with 76 percent accuracy
A startup called Sunu has developed a wrist-worn device called the Sunu Band to help people with visual impairments. The device uses sonar to automatically detect when it is close to an obstacle and vibrate to alert the wearer that he or she might run into it. The Sunu Band can vibrate with varying intensity to let the wearer know how close an object is, growing stronger when an obstacle is in his or her immediate proximity.