This week’s list of data news highlights covers April 7-13, 2018, and includes articles about how Dubai is testing smart license plates and a new satellite that will be able to spot methane leaks from space.
The Japanese government has finalized a plan to make anonymized patient data available to researchers and pharmaceutical companies. The plan, which will go into effect on May 11, 2018, allows medical institutions to pool consenting patients’ medical records, submit the data to a government-approved firm to anonymize the data, and then make it available for academic and private sector research.
Dubai’s Roads and Transport Authority is launching a trial of license plates equipped with digital screens, GPS, and communication ability. The license plates are designed to report a vehicle’s location to emergency services in the event of a crash, share data about traffic conditions with nearby vehicles, and display an alert if a vehicle is reported as stolen.
Researchers at the nonprofit Gladstone Institutes, the University of California, San Francisco, and Google have developed a machine learning system that can analyze neuronal cells, which could allow scientists to automate the labor-intensive process of manually labeling neurons and studying their changes over time. To study brain cells, scientists typically need to label the cells with fluorescent dyes, which can damage the cells, and then compare microscopic images to track their changes, which requires large amounts of time and extreme attention to detail. The researchers’ system can identify specific types of neurons in new images, distinguish between live and dead cells, and differentiate between cell parts, all without the need for fluorescent labels.
The U.S. Food and Drug Administration (FDA) has approved an AI system called IDx-DR for use in diagnosing an eye disease called diabetic retinopathy without the need for a clinician. The system has an error rate of about 13 percent, compared to human ophthalmologists, which have an error rate of between 20 and 30 percent when diagnosing the disease. IDx-DR is the first diagnostic software that the FDA has approved for use without a medical specialist, allowing primary care physicians and other healthcare professionals to use it without the need for patients to visit an ophthalmologist.
Researchers at the University of Washington have developed an AI system that acts like a dog by training on data about dog behavior. The researchers outfitted a dog with a camera and motion sensors on its legs, tail, and body, to create a dataset of 24,500 video frames synchronized with movement and body position data. The researchers trained their system on 21,000 of these frames, and when testing it on the remaining frames, demonstrated that it could accurately predict how a dog would move, plan to move between different locations, and differentiate between walkable and non-walkable surfaces.
Environmental advocacy group Environmental Defense Fund (EDF) has announced that it will develop and launch a satellite called MethaneSAT by 2021 capable of detecting methane leaks. While existing satellites can detect methane on Earth, they cannot do so with high enough resolution to identify specifically where a leak is coming from. EDF plans to use MethaneSAT to monitor 50 major oil and gas producing regions, which account for over 80 percent of global methane production, and make this data freely available to the public.
China has issued national guidelines for testing self-driving cars designed to make the country competitive with the United States and other leaders in autonomous vehicle research. The guidelines require self-driving vehicle tests to initially take place off of public roads, but companies can then test their systems on designated public roads provided they have a human driver ready to take over.
Researchers at the University of California, Berkeley and the University of British Columbia have developed an AI system capable of learning from video footage of people performing martial arts and acrobatics and training a 3D model to imitate their actions. The system uses real world examples as a baseline and then uses reinforcement learning for the model to practice the actions with realistic physics until it gets them right.
Google has published two tools to demonstrate advancements in a natural language processing technique called word vectors that allow AI to intuitively interpret language. The first tool is called Talk to Books and allows users to ask a question or make a statement, and it will automatically identify a relevant response in a book without using keyword matching. The second is a word association game called Semantris that challenges users to predict strongly related words, including similar concepts, related concepts, and antonyms.
The U.S. Department of Agriculture (USDA) has resumed publishing detailed animal welfare data as open data after blocking public access to this information in February 2017. The USDA’s Animal and Plant Health Inspection Service (APHIS) database made reports about animal welfare inspections and USDA enforcement actions freely available to businesses and the public until USDA abruptly restricted access to it. In response to public outcry, USDA began publishing limited amounts of this data months later, causing many to rely on Freedom of Information Act requests to access this information, which can take months. USDA finally resumed making APHIS data fully available after Congress publicly criticized the agency for needlessly limiting the data it published.