This week’s list of data news highlights covers May 11-17, 2019, and includes articles about an AI system predicting heart attacks and an AI robot that can pick tomatoes without bruising them.
1. Tracking Bad Actors on the Dark Web
Researchers from MIT have developed an AI system that can identify if users from different forums on the dark web are the same individual. It is difficult for law enforcement to track individuals using the dark web because of the fast rate individuals create and close dark web marketplaces. The researchers’ system correctly identifies a match 95 percent of the time by analyzing similarities in usernames, in content users write, including similar phrases, and in the community with which users interact.
Researchers from Finnish and Dutch medical centers have shown that a machine learning model can accurately predict if patients with chest pain are at risk of suffering heart attacks or dying. Researchers provided the model data on 85 variables, including age, hypertension, and blood flow, concerning 950 patients with heart pain. The model then predicted which patients suffered heart attacks or died over a six-year period with 90 percent accuracy.
3. Making TV Captioning More Accurate
Researchers from IBM have developed an AI system that can transcribe broadcast news captions almost as accurately as humans. The researchers trained the system on more than 1,300 hours of broadcast news audio, helping it learn to identify words from overlapping speech and from individuals with a variety of speaking styles. On two test datasets, the system had error rates of 6.5 percent and 5.9 percent, compared to human error rates of 3.6 percent and 2.8 percent, respectively.
4. Predicting if a Patient Will Respond to Treatment
Researchers from the University of Michigan and the U.S. Department of Veteran Affairs have developed a machine learning model that accurately predicts if patients with Crohn’s disease will respond to treatment. The model analyzed the demographic and laboratory data-—including a person’s age and hemoglobin level—of 400 patients, to make predictions. The model could help bring down healthcare costs for patients by reducing the use of drugs that would likely prove ineffective.
Researchers from MIT and Washington University in St. Louis have shown that eBay’s AI translation tool, which translates the titles of listings, increased the firm’s sales from U.S. sellers to individuals in Spanish-speaking Latin American countries by more than 10 percent. The researchers compared sales data before and after eBay implemented the tool in 2014, and accounted for other variables such as increased marketing spend and the length of title listings. The researchers found similar increases in sales after eBay implemented AI translation tools for Russian, Italian, and French.
6. Hearing What You Want to Hear
Researchers from Columbia University have developed an AI system that can amplify specific voices in a noisy room based on who a listener wants to hear. The brain produces electrical signatures when a person is listening to something, and the researchers trained a deep learning algorithm to analyze the brainwaves of listeners while they listen to different sound sources to determine the sound the individual wants to hear. The system could help hearing-impaired individuals.
7. Prolonging Independence for the Elderly
TigerPlace, an intermediate-care senior community in Missouri, has used a system of sensors to extend the number of years their residents can maintain independence. Depth sensors create 3D silhouettes of individuals moving in their apartments, allowing the medical staff to track residents’ walking speed, stride length, and time spent walking. In addition, bed sensors measure residents’ heart rate, respiration rate, and restlessness. This data has helped the medical staff identify early signs of health changes in residents, including using data from bed sensors to identify stiffness in residents’ arteries.
8. Making it Easier to Detect Ear Infections
Researchers from the University of Washington have developed a smartphone-based system that can detect the signs of ear infections. Individuals using the system fold a piece of paper into a funnel shape and tape it around a phone’s microphone and speakers, and the researcher’s app then produces birdlike chirping sounds. The phone’s microphone detects the sound waves bouncing off the eardrum and the app analyzes the echo to determine if it matches the vibrations of healthy eardrums.
9. Building an AI Robot That Can Pick Tomatoes Without Bruising Them
Root AI, a Massachusetts-based startup, has developed an AI-powered robot that can pick tomatoes and detect ripeness better than humans. The robot uses cameras and sensors to navigate and has a dexterous hand that picks tomatoes without bruising them or tearing down vines. Autonomous produce-picking robots could help address the labor shortages in the United States’ agricultural industry.
Researchers from Google have created a new AI translation system that can provide translations faster while preserving the vocal characteristics of the original speaker. Most translation systems first transcribe speech, then translate the transcribed speech into text of the target language, and then generate speech in the target language. The researchers’ system does not rely on an intermediate text representation of the speech in either language, allowing the system to translate faster and making it easier for the translated audio to retain the voice of the original speaker.
Image: PxHere
