10 Bits: the Data News Hotlist
This week’s list of data news highlights covers August 27 – September 2, 2016 and includes articles about a new partnership to advance self-driving cars and a wearable device that can diagnose voice disorders.
The New York State Department of Motor Vehicles (DMV) has caught 100 identity thieves and opened 900 new cases by improving its facial recognition algorithm that analyze people’s faces when they apply for a driver’s license. When someone applies for a driver’s license in New York, the DMV’s facial recognition algorithm analyzes the applicant’s photograph and compares it against a 16 million photograph database to ensure the person is not lying about his or her identity. By doubling the number of measurement points the algorithm analyzes, from 64 to 128, the DMV was able to substantially improve the system’s accuracy to catch identity thieves.
The United Kingdom’s National Health Service (NHS) has partnered with Google DeepMind, an AI research division of Alphabet, to use machine learning to improve how radiotherapists analyze medical scans to treat patients with head and neck cancers. Radiotherapists scan patients and rely on a technique called segmentation to differentiate between healthy tissue and cancerous tissue to target with radiation, however segmentation can take several hours for head and neck cancers. DeepMind will analyze anonymized scans from 700 patients to train machine learning segmentation algorithms, with the goal of reducing the time the process takes to just one hour.
Baidu has partnered with Nvidia to develop an AI platform for self-driving cars with the goal of launching a self-driving taxi service in China. Baidu will take advantage of Nvidia’s existing self-driving AI technology, which includes a supercomputer designed specifically to analyze data from vehicle sensors and advanced mapping software, and integrate it into its ongoing autonomous vehicle research. Baidu plans to launch a self-driving shuttle service in China by the end of 2018 and bring self-driving cars to 10 cities by 2019.
The Allen Institute for Artificial intelligence (Ai2), a research institute based in Seattle, have developed an AI system that can help machines make better predictions about the physical properties of different scenery. Researchers at Ai2 created 3D modeled scenes based on 10,000 different images and used a 3D physics engine to estimate the physical properties of different objects in a scene. The researchers then trained an artificial neural network with these scenes to teach it to estimate the physical characteristics of objects, such as how they would respond to movement. After this training, the system could accurately predict the physical characteristics of unfamiliar objects, which could eventually improve how systems like self driving cars respond to unfamiliar environments.
Researchers at the Massachusetts Institute of Technology and Massachusetts General Hospital have developed a wearable device equipped with an accelerometer that collects data about the movement of a user’s vocal cords and uses machine learning to determine abnormal speech. People with muscle tension dysphonia (MTD), a voice disorder, experience vocal fatigue and deteriorating voice quality, despite not having physically damaged their vocal cords, which makes the condition hard to diagnose and treat. The wearable sensor detects the subtle movements of vocal cords and connects to a smartphone, where machine learning algorithms can detect signs of MTD. The system could be particularly useful for collecting vocal performance data to aid diagnostics without having patients spend much time with specialists in person.
Tesla has updated its partially autonomous autopilot system to automatically detect when a driver is dangerously misusing the feature and force the driver to retake manual control of the car. Autopilot requires drivers to keep their hands on the steering wheel while the feature is active, and if the car senses a driver has removed his or her hands, it will warn the driver. If the driver does not comply after the warning, Autopilot will disengage and the driver will not be able to turn the feature back on until he or she parks the car.
The U.S. National Highway Traffic Safety Administration (NHTSA) has launched an initiative to use open data to help reduce traffic fatalities after 2015 saw a 7.2 percent increase in fatal accidents from 2014. NHTSA will publish anonymized records about all traffic accidents it recorded for 2015 as open data, several months ahead of its regular publication date. It is also soliciting public feedback about the data and solutions to reduce traffic accidents. Several companies have committed to support the initiative as well, such as navigation app Waze, which will share real-time traffic data with the Department of Transportation, and mapping software company Mapbox, which will develop tools to help inform drivers about fatal accidents nearby.
Baidu has made its machine learning platform PaddlePaddle freely available as open source, following the growing trend of major technology companies such as Google, Microsoft, Facebook, and Amazon that are opening up their AI tools to the public. PaddlePaddle, which focuses on deep learning applications, is reportedly more user-friendly than other open source AI tools, and Baidu hopes making it publicly available will encourage more AI research. An early version of PaddlePaddle is available on GitHub, and the full version will be available on September 30th.
The U.S. Environmental Protection Agency (EPA) has launched the Smart City Air Challenge to encourage the development of networked air sensor technologies that can effectively monitor pollution in cities. The challenge calls for communities to develop data collection, management, and sharing systems that can support hundreds of connected sensors and the winning community will receive funding to implement their systems. Winners must also share their technologies with EPA and other communities at the end of the challenge to encourage increased deployment of air sensor networks.
A small family cucumber farm in Japan has developed a machine powered by Google’s open source TensowFlow machine learning platform that can distinguish between different varieties of cucumber and sort produce more efficiently than humans. The farm sorts cucumbers by size, shape, color, and other characteristics, which can be a difficult and time-consuming process. To simplify this process, a farmer and former embedded systems designer built a tool utilizing TensorFlow by training it on 7,000 annotated images of cucumbers so it could learn to distinguish between them and separate them with a mechanical arm. Now, the farm uses the system to automatically sort cucumbers with 70 percent accuracy instead of human labor.