This week’s list of data news highlights covers February 6, 2021 – February 12, 2021 and includes articles about optimizing the design of renewable energy systems and influencing human decision-making with AI.
Researchers from the Tokyo University of Science and the National Cancer Center Hospital in Japan are using machine learning and electromagnetic radiation to identify gastrointestinal stromal tumors (GISTs), which are difficult to detect because they are often covered by a layer of connective tissue. The team trained an algorithm to analyze and identify GISTs using images of tissues a pathologist had tagged as either normal or tumor tissue. When tested, the algorithm correctly identified ten out of twelve tumors that were covered by a tissue layer, and it was able to color-code tumorous and non-tumorous sections with 86 percent accuracy.
The U.S. Department of the Treasury is piloting the use of an AI algorithm to extract the most pertinent information from PDF files of Congressional spending bills. The algorithm identifies useful information, such as which agencies are to receive funding and how much, using natural language processing to understand the syntax and sentence structure within bills, and extracts this information for a bureau expert to examine and flag for any inaccuracies. This can streamline the annual appropriations process and expedite when government agencies receive their money.
Researchers at IBM have used an AI system to analyze 278 million social media posts to find which Super Bowl advertisements best motivated viewers to buy their products. The system found that the advertisement for Cheetos (a brand of artificial cheese puffs) generated the most purchase motivation for viewers based on the number of mentions and social media posts that indicated an intent to purchase. In general, the online conversation around the Super Bowl in 2021 focused mostly on snacks, compared to last year, which focused heavily on music and celebrities.
Researchers from Qassim University in Saudi Arabia, the Minia University, and the Aswan University in Egypt, have developed an algorithm that optimizes the design of renewable energy systems. The algorithmic model is called Improved Artificial Ecosystem Optimization (IAEO) and works by finding which combination of different kinds of renewable energy sources yields the most energy for the lowest cost. The researchers applied IAEO to six different green energy system designs, and found that the energy system designs IAEO chose were more efficient than those chosen by existing models.
Researchers from UT Southwestern Medical Center in Texas have developed an algorithm that estimated that the number of people infected with COVID-19 in the United States is almost three times higher than confirmed cases. The algorithm’s predictions are derived from the number of reported coronavirus-related deaths, rather than the amount of lab-confirmed cases. It assumes that the infection fatality rate is 0.66 percent, based on early pandemic data in China, and incorporates other factors, such as the average time it takes for someone with symptoms to either die or recover. When tested, the researchers found that the algorithm’s estimates correspond closely with studies quantifying the number of people who have tested positive for SARS-CoV-2 antibodies.
Researchers from Cornell University have developed an AI algorithm that can identify physical human interactions using just a person’s shadow. Researchers trained an algorithm to distinguish between shadow images of people performing six gestures (touching with a palm, punching, touching with two hands, hugging, pointing, and not touching). When tested, the algorithm accurately identified gestures under daylight with 96 percent accuracy. With this algorithm, the team envisions teaching robots to better respond to more gestures.
Researchers from the Australian National University and Commonwealth Scientific and Industrial Research Organisation have led a study that found that AI systems can influence human decision-making. In a set of experimental games, participants clicked on colored boxes to win fake currency and were prompted to choose between two investment options. The participants were acting as investors and the AI system was acting as the trustee. The researchers found that the AI system would guide players towards specific investment choices after learning their habits and choice patterns. After several rounds of the game, the AI system had learned how to get participants to give it more money.
Researchers from Google and the U.S. Centers for Disease and Control and Prevention have combined machine learning technology with smartphone tracking data to forecast the spread of the flu. The researchers trained a machine learning system to recognize human movement on a city map using anonymized tracking data from Android phone users. The team then added in data about influenza-related patient hospital visits and lab reports, which the system used to forecast the spread of influenza based on human movement. When the researchers used the system to predict the 2016 influenza season in Australia, they found that their system accurately tracked the spread of an outbreak across international lines, unlike commuter data-based systems.
NASA and Hewlett Packard Enterprise have created Spaceborne Computer-2, the first commercial computer to operate in space with enough processing power to support AI and machine learning capabilities. Spaceborne Computer-2 is capable of processing image data that space stations, satellites, cameras and other sensors receive, eliminating the need to send data back to earth for ground-based processing, a process that can take several months.
Japan’s Mitsui O.S.K. shipping lines have collaborated with Bearing, a U.S. AI technology start-up focused on improving maritime shipping operations, to achieve greater routing efficiency to lower carbon emissions using AI. The AI models use voyage data, such as vessel speed, trim, main engine operation, weather, and sea condition, to predict a vessel’s fuel consumption. Using this information, the model analyzes potential routes for a voyage and recommends the route with the most optimal energy output.
Image credit: Ed White