This week’s list of data news highlights covers March 6, 2021 – March 12, 2021 and includes articles about combining soil data to improve crop growing techniques and using AI to help create cleaning schedules in airports.
Scientists from the University of Oxford and the UK Center for Ecology and Hydrology have used pattern recognition and machine learning to identify the amount of raw sewage that has leaked into UK rivers from two wastewater treatment plants. The team trained an algorithm to identify the difference in flow rates for treated and untreated wastewater using 11 years of data. The algorithm identified that there have been 926 leaks where untreated sewage has flowed for at least 3 hours.
IBM’s research team is developing new antibiotics using AI. First, the researchers used an algorithm to examine the structure and sequence of amino acids in existing drugs, which defines how these drugs function. Then, they used this data to develop new antibiotics with specific, desired properties including the ability to kill bacteria, work across a range of different classes of bacteria, and be safe for human use. Using these algorithms, IBM has developed two new antibiotic candidates that had low toxicity when tested in mice and cell cultures, and proved unlikely to lead to drug resistance in E. coli.
Lightmatter, a photonics start-up based in Boston, has developed a computer chip that uses light to perform calculations. Conventional computer chips use transistors to control the flow of electrons and perform a wide variety of calculations, but Lightmatter’s chip manipulates laser beams to perform specific types of mathematical calculations that are critical to AI computation. When running a natural language model, Lightmatter’s chip was five times faster and consumed one-sixth of the power a conventional AI chip uses.
The Riken Center for Computational Science and Fujitsu Laboratories, the research and development division of the Japanese IT company, has officially completed building Fugaku, the world’s fastest supercomputer, after seven years. Japan has selected 74 research projects to use Fugaku for various applications, such as simulating drug discovery processes and creating new materials for batteries and fuel cells.
Grow Observatory, an EU-funded platform that combines low-cost consumer sensing technology to combat climate change, has created a network of 6,500 ground-based soil sensors in 13 European countries. The insights from this data provide location-specific crop and planting advice, helping people test regenerative food growing techniques. For example, local farmers in the Canary Islands have been able to reduce their use of water for irrigation by 30 percent using information on soil moisture levels from the Grow Observatory app.
Graduate students at Carnegie Mellon University in Pennsylvania are developing an AI system to create smarter cleaning schedules for custodial staff at airport restrooms. The system will use data on flight arrivals and sensor data on the number of people entering a restroom, how often soap dispensers become empty, and how quickly garbages fill up to best allocate staff to the restrooms that need to be cleaned. After being developed, the system will be piloted at Pittsburgh International Airport, which has experienced limited staff capacity due to COVID-19.
Manufacturing companies are using real-time data to improve decision-making and lower costs. SilencerCo, a manufacturer of firearm suppressors in Utah, has used equipment data and manufacturing times to increase the utilization of machines by 8 percent, improve the production of sellable parts by 200 percent, and eliminate 11,500 hours of machine downtime. BC Machining, a manufacturer in North Carolina that provides aluminum machining services, has used equipment data to develop an algorithm that predicts when a machine will fail. Since being implemented, BC Machining has eliminated nearly 100 percent of scrap parts and saved $72,000 on machine maintenance per year.
Researchers at the University of Washington have developed a set of machine learning algorithms that can identify and monitor heartbeats without any physical contact. The algorithms identify heartbeats using a smart speaker that sits within two feet of a person and sends out inaudible sounds that bounce off them and return back to the speaker. The algorithms then recognize whether a heartbeat is regular or irregular depending on how the sounds are reflected.
Waymo, an U.S. autonomous driving company, has released an open dataset of motion videos captured by sensors on vehicles. The dataset covers 6 urban cities and includes 100,000 video clips of objects, such as cars and people, and their trajectories. The dataset also contains 3D maps, which researchers can use for prediction modeling, and annotations of perception boxes, which track objects in motion to help developers build more accurate prediction models.
Researchers at the University of Washington have developed a machine learning model that can predict which patients with asthma are more likely to have poor outcomes in the future and explain how it came to its conclusion. Previous models were accurate but did not offer explanations as to why patients were at risk. The researchers’ new model automatically explains its prediction results without lowering the accuracy of the model’s performance. When tested, the model explained its predictions for around 90 percent of patients who were correctly predicted to need hospital visits for their asthma in the following year.
Image credit: Free-Photos