This week’s list of data news highlights covers August 25 – September 3, 2018, and includes articles about a GPS system for space navigation and a startup developing an address system for the entire world.
Researchers at the Massachusetts Institute of Technology and Columbia University have developed an AI tool for computer-aided design software that can optimize the design of an object for factors like drag and stress tolerance. Designers typically use an iterative process for their designs, creating many versions of each and adjusting them to make improvements. The AI tool can automatically generate dozens of variations of a design that each optimize different factors, bypassing the need for iterative design processes, which can be time-intensive and require multiple human designers.
Researchers at NASA and Intel are developing a new kind of global positioning system (GPS) to help astronauts and rovers navigate space that uses AI to interpret images of its surroundings, rather than rely on satellite triangulation, which is not possible on other planets without satellites. The GPS uses images of its surroundings and a database of millions of images of a particular area to identify its location relative to other images. The researchers tested the system by giving the GPS 2.4 million images of a simulated moon and instructing it to identify a specific location in a new image of the moon, demonstrating it could effectively navigate the surface of the virtual moon.
Two teams of professional human gamers managed to beat a team of AI bots from AI research nonprofit OpenAI, called the OpenAI Five, in the video game DOTA 2, ending the OpenAI Five’s winning streak against human players. Developing AI that can beat humans at highly complex games is a major goal for AI researchers, as it requires AI systems to learn how to cooperate, strategize, and learn and respond to creativity. Though the OpenAI Five lost, the matchups demonstrated the AI’s significant progress and highlight opportunities for AI researchers to focus on improving how AI can develop long-term strategy.
Researchers at the University of Central Florida have developed an AI system that can detect small lung cancer tumors that humans struggle to identify in computerized tomography (CT) scans. The researchers trained their system on over 1,000 CT scans and in a test, the system could identify small lung cancer tumors with 95 percent accuracy, whereas humans are only 65 percent accurate.
A London-based mapping startup called What3Words has developed an address system for the entire world that uses combinations of 3 words for each location. Traditional address systems can be confusing and vary substantially between countries, and developing countries may lack these systems entirely, posing challenges for businesses, mail systems, and emergency services. What3Words assigns randomized 3 word combinations, such as “lakefront.boundless.vitals,” to every 3 square meter block in the world, approximately 57 trillion squares in all.
Researchers at the Institut Gustave Roussy, a French cancer research institute, has developed an AI system that can predict how effective immunotherapy will be for a patient, which could help reduce treatment time and increase chances for success. The system analyzes CT scans of tumors to assess the concentration of lymphocytes, which are white blood cells that are part of the immune system. The researchers had their system analyze CT scans and data about patient outcomes and found that the greater the concentration of lymphocytes, the more likely a particular type of immunotherapy would be effective, allowing the system to create a score indicating the likelihood of success.
Researchers at Harvard University have developed a machine learning system that can predict the location of aftershocks after an earthquake more accurately than traditional methods. Aftershocks can be even more damaging than the initial earthquake and can occur months later, and while scientists can typically predict how large aftershocks will be, they cannot always identify where they will occur. The researchers trained their system on data about 131,000 earthquakes and aftershocks and in a test could predict the location of aftershocks more accurately than the current leading approach.
Pharmaceutical companies are increasingly turning to clinical trials that involve in-home data collection and mobile reporting, which could increase clinical trial participation. A life sciences company called AOBiome Therapeutics successfully completed a 12-week clinical trial of an acne drug by mailing participants the drug and having participants use a smartphone app to record their experiences and regularly submit photos of their condition. Similarly, French pharmaceutical company Sanofi has completed a trial of a drug that involved participants using sensors and wireless devices to log their own weight, mobility, blood pressure, and blood glucose levels at home. Current clinical trials struggle to find and retain volunteers as they often involve large time commitments at treatment facilities.
X-ray detector manufacturer Smiths Detection has developed an AI system that can allow its machines to spot lithium-ion batteries in luggage with up to 90 percent accuracy. Many aviation authorities prevent people from bringing certain kinds of batteries, which are commonly used in smartphones, photography equipment, and other devices, in checked baggage as there is a risk they could short-circuit and explode. Smiths Detections’ system could make it easier and quicker for airports to scan checked baggage.
Researchers at Saarland University in Germany and the University of South Australia have developed an AI system that can track a person’s eye movements to predict certain personality traits such as sociability and conscientiousness. The researchers had 50 volunteers wear an eye-tracking device and walk around a university campus and buy something at a store and then complete a questionnaire about their personality. Then, the researchers used AI to establish links between reported personality traits and over 200 different markers from the eye-tracking data, such as blinking frequency and pupil dilation. The researchers hope to improve the system by teaching it to incorporate data about body language, and believe it could eventually be useful for people with autism by helping them better interpret others’ behavior.
Image: Leggi il Firenzepost.