Published on May 12th, 2017 | by Joshua New0
10 Bits: the Data News Hotlist
This week’s list of data news highlights covers May 6-12, 2017, and includes articles about a new tool to help the homeless in New York City and Canada’s new data analytics research institute studying wireless spectrum.
Researchers at Ludwig Maximilian University of Munich have developed a machine learning method capable of relating how 1,169 different languages use verb tenses, which could lead to more effective machine translation for uncommon languages. The researchers built a database of 1, 169 translations of the Bible and manually annotated conventions that indicate verb tense in 10 of these languages, such as how “is” and “was” indicate present and past tense in English. Then the researchers trained the system on these annotated languages than then had it analyze the entire dataset to identify similar conventions in every language, effectively creating a relational map of each language.
New York City’s Human Resources Administration has launched a new tool called StreetSmart that will serve as an analytics platform and dashboard for city workers and nonprofits working to reduce homelessness in the city, which has reached rates not seen since the Great Depression. StreetSmart will replace the multiple siloed systems different organizations use, allowing caseworkers to upload the data they regularly collect, such as health, income, and demographic data, share casefiles, and track individual cases over time. StreetSmart also includes mapping tools to allow caseworks to better understand where homeless encampments arise and how effective different approaches, such as developing new housing facilities, are at reducing homelessness in an area.
Loveflutter, a startup online dating service in the United Kingdom, has partnered with sentiment analysis firm Recptiviti AI to develop a tool called Analyze140 that analyzes users’ tweets and matches them with people with complementary personalities. Analyze140 can generate scores for a users’ emotional and social characteristics based on factors such as word choice and writing style, and Loveflutter uses these metrics to determine how compatible different users are.
Researchers at the Research Center for Brain-Inspired Intelligence in Beijing have developed a deep learning system that can analyze functional magnetic resonance imaging (fMRI) scans of the brain’s visual cortex and reconstruct images it perceives. fMRI scans can be difficult for computers to analyze because the voxels—the equivalent of a pixel in a 3D model—it generates to represent brain activity contain large amounts of noisy data. The researchers trained their deep learning system on 1,800 fMRI scans and images a subject was viewing while they were scanned to teach it to identify the relationship between voxel representation and the original image, and then recreate the image with a high degree of accuracy.
The Canadian government has launched a research organization called the Big Data Analytics Centre to collect and analyze data on Canada’s radiofrequency spectrum and guide spectrum policy to ensure Canada can fully take advantage of connected technologies. The Big Data Analytics Centre will be based in the government’s Communications Research Centre in Ottowa.
University of Nevada researchers have developed a four-wheeled autonomous robot named Seekur that uses using sensors and machine learning to analyze the structural integrity of bridges. Seekur is capable of navigating itself across a bridge while ground-penetrating radar, a camera, and corrosion-detecting sensors analyze aspects of the bridge’s health, while a machine learning algorithm converts this data into a heat map of the bridge flagging weak points for human engineers. In tests on bridges in several states, Seekur proved to be faster and more accurate than humans.
The winning team of Kaggle’s annual Data Science Bowl competition have successfully developed an artificial neural network capable of analyzing low-dose computed tomography (CT) scans to detect signs of lung cancer. Low-dose CT scans are desirable because they use smaller amounts of radiation and do not require patients to be injected with contrast dye, though they produce a high rates of false positives The team, from Tsinghua University in China, were able to develop their system by training it on the 2,000 scan dataset provided by Kaggle and teaching it to identify potential signs of lung cancer and then estimate the probability that a patient actually has lung cancer, which could help reduce the number of false positives and unnecessary medical procedures involved in lung cancer diagnoses.
The New York Police Department (N.Y.P.D.) has developed an updated version of its crime-fighting analytics platform CompStat to carry out and analyze daily surveys of New York City residents about their feelings about the police. The system regularly sends surveys to 50,000 volunteers’ smartphones asking about how safe they feel, as well as their trust and confidence in N.Y.P.D., and turns this data into a series of scores. Precincts can view these scores to compare them against crime statistics and gauge the impact of different policing strategies on public sentiment.
Researchers at Salesforce have developed a machine learning algorithm capable of summarizing long pieces of text into concise summaries with a high degree of accuracy. The algorithm combines several different machine learning techniques, including supervised learning and reinforcement learning, to identify important points in a body of text, as well as uses a mechanism that allows it to pay particular attention to text based on its context, which allows it to avoid repeating itself, which often challenges summarization algorithms.
A team of biomedical engineers at the Swiss Federal Institute of Technology in Lausanne has developed a wearable robotic exoskeleton called the Active Pelvis Orthosis (APO) designed for the elderly that uses sensors to quickly detect if a wearer begins to slip and provide additional support to stop them from falling. APO can learn a wearer’s walk in three minutes and can identify sudden changes indicating a fall within 350 milliseconds, signaling it to rotate a wearer’s hips in a way that prevents him or her from falling.