Welcome to the February 16, 2022, edition of ACM TechNews, providing timely information for IT professionals three times a week.
ACM TechNews mobile apps are available for Android phones and tablets (click here) and for iPhones (click here) and iPads (click here).
To view "Headlines At A Glance," hit the link labeled "Click here to view this online" found at the top of the page in the html version.
The online version now has a button at the top labeled "Show Headlines."
|
|
|
Good News About the Carbon Footprint of ML Training
Google AI Blog
David Patterson
February 15, 2022
Scientists at Google and the University of California, Berkeley, investigating how to reduce the energy and carbon footprint of machine learning (ML) training of natural language processing models, identified four best practices that together can cut energy and carbon emissions “significantly.” According to David Patterson, ACM A.M. Turing Award recipient and a Distinguished Engineer at Google Research, these model, machine, mechanization, and map optimization (4M) practices are based on selecting more-efficient ML model architectures, and employing ML-optimized processors and systems to enhance performance and energy efficiency. Other helpful practices include reducing emissions through the use of cloud computing, and selecting locations for datacenters that offer the cleanest energy. Patterson said the 4M practices together can reduce the energy requirements of computing in datacenters by 100 times, and emissions by 1,000 times.
|
Advanced Computer Simulations Shed Light on Magma Below Earth's Surface
University of Bristol News (U.K.)
February 14, 2022
An international team of researchers led by the U.K.'s University of Bristol used computer modeling to explore magma deep within the Earth. The researchers used the U.K.'s ARCHER national supercomputing service to simulate the physical properties of magma at the transition zone between Earth's upper and lower mantle, at temperatures up to 1,600 degrees Celsius (2,912 degrees Fahrenheit) and 250,000 times atmospheric pressure. Analysis revealed that water-rich magmas formed in the mantle are more buoyant and fluid than previously thought. Said Bristol's James Drewitt, "Using advanced computational techniques to model hydrous magmas down to the atomic scale, we discovered natural hydrous magmas will be more buoyant and fluid than expected, and will therefore rise through the upper mantle towards the surface, rather like the wax rising in a lava lamp.”
|
An Atlas of Cells That Carry Blood to the Brain
MIT News
Anne Trafton
February 14, 2022
Researchers at the Massachusetts Institute of Technology (MIT) have produced a comprehensive atlas of human cerebrovascular cell types and their functions, which can potentially help in the discovery of new drug targets for Huntington’s disease. The researchers acquired postmortem and healthy brain tissue samples, and conducted single-cell RNA-sequencing on over 16,000 cerebrovascular cells, broken down into 11 subtypes. Analysis revealed evidence of zonation, in which the endothelial cells that line blood vessels express different genes based on their location. "Our goal is to build a systematic single-cell map to navigate brain function in health, disease, and aging across thousands of human brain samples," said MIT’s Manolis Kellis. "This study is one of the first bite-sized pieces of this atlas, looking at 0.3% of cells."
|
Computer Models Show How Crop Production Increases Soil Nitrous Oxide Emissions
Iowa State University News Service
February 15, 2022
Iowa State University (ISU) scientists used computer models to determine the increase in emissions of greenhouse gas nitrous oxide over the last century by U.S. crop production. The researchers factored in government data on crops, land use, weather, and other variables, as well as historic and survey data from growers and other landowners. ISU's Chaoqun Lu said, "We divide land into thousands of pixels at a uniform size and run algorithms that simulate how ecological processes respond to changes in climate, air composition, and human activities." The analysis found nitrous oxide emissions from U.S. soil have risen more than threefold since 1900, from 133 million metric tons of carbon dioxide equivalent (MMT CO2 eq) annually to 404 MMT CO2 eq annually in the 2010s. They found nearly two-thirds of that expansion is rooted in agricultural soils, with corn and soybean production fueling over 90% of the agriculture-related emissions increase.
|
Remapping Atrial Fibrillation Treatment
UC San Diego Health
Corey Levitan
February 14, 2022
Researchers at the University of California San Diego Health (UC San Diego Health) have developed a non-invasive computational mapping system to help treat atrial fibrillation and other heart arrhythmias. The vMap system generates a three-dimensional (3D) interactive map of arrhythmia hotspots in the heart in just minutes, using data from a standard 12-lead electrocardiogram. Said UC San Diego Health’s Dr. David Krummen, “We designed vMap to improve ablation (treatment) outcomes by rapidly providing arrhythmia source information to the physician. We want this technology to increase first-pass ablation success, decrease procedural risk, and improve the care of patients with heart rhythm abnormalities." vMap was approved by the U.S. Food and Drug Administration in November.
|
Genetic Database to Identify Missing Persons in El Salvador
Universitat Pompeu Fabra Barcelona (Spain)
February 14, 2022
Researchers at Spain's Universitat Pompeu Fabra Barcelona and the Institute of Evolutionary Biology have developed a genetic database that aims to identify people who went missing or were put up for adoption illegally during the 1980-1992 war in El Salvador. The database contains the genetic information of 400 Salvadorans, which also could help identify the remains of missing persons along the migrant route into the U.S. Said researcher Ferran Casals, "We have developed pioneering applications of the most advanced nucleic acid sequencing technologies in forensic genetics." Researcher Francesc Calafell added, "Thanks to the good resolution of the genomic markers sequenced, and thanks to the large number of sequences of individuals accumulated, the database has allowed us to go a little further and not just identify parents and children, but also more distant relations."
|
Wearable Sensors for COVID-19 Mitigation
News-Medical Life Sciences
Neha Mathur
February 13, 2022
Researchers in Canada and the U.S. have demonstrated the potential of digital contact tracing technology in mitigating the COVID-19 pandemic by modeling wearable sensor deployments. The team designed a compartmental epidemiological model based on a Susceptible, Exposed, Infectious, Removed (SEIR) framework, in which wearables alerted users to potential infection and urged them to seek laboratory-based tests and self-quarantine. Researchers simulated wearable sensor deployment scenarios during Canada's second COVID-19 pandemic wave; the results suggested deployment eased the burden of infection by shrinking the pool of infectious individuals, although this was heavily dependent on detection algorithm specificity. The researchers concluded that wearable deployment is generally more beneficial than broad antigen test-based screening approaches.
|
Researchers Train Neural Network to Recognize Chemical Formulas From Research Papers
Skolkovo Institute of Science and Technology (Russia)
February 14, 2022
Scientists at Russia's Syntelly automation startup, Lomonosov Moscow State University, and the Sirius University of Science and Technology have educated neural networks to automatically identify chemical formulas in research papers. The researchers used Google's Transformer machine translation neural network to convert images of molecules or molecular templates into textual representations named Functional-Group-SMILES. The network was able to learn almost anything it was provided, as long as the pertinent depiction style was represented in the training data. The researchers also designed a data generator to produce examples of molecular templates by blending randomly chosen molecule fragments and depiction styles. Syntelly's Sergey Sosnin said, "Our study is a good demonstration of the ongoing paradigm shift in the optical recognition of chemical structures.”
|
Satellite Imagery Gives Researchers Timeline of When Swine Waste Lagoons Were Built
NC State University News
Matt Shipman
February 14, 2022
An automated technique developed by North Carolina State University (NC State) researchers taps satellite imagery to ascertain when swine waste lagoons were built, so scientists can trace their impact on environmental quality. The project began with location data on 3,405 swine waste lagoons; the researchers then gathered publicly available satellite images of the regions in which the lagoons are sited. They automated a piece of software to learn each lagoon's construction timeline, and found “that approximately 16% of the lagoons were already in place before 1987,” said NC State’s Lise Montefiore. In addition, she said, “We were able to establish the construction date for the remaining lagoons with a margin of error of about one year.”
|
Quantum Errors Made More Tolerable
ETH Zurich (Switzerland)
Andreas Trabesinger
February 11, 2022
Physicists at the Swiss Federal Institute of Technology, Zurich (ETH Zurich) have demonstrated the ability to extend the longevity of quantum states and expand tolerance of quantum errors, which are crucial to future quantum computing. The method accounts for limitations of physically realistic devices, and is relatively easy to deploy compared to other proposed error-correction schemes. The researchers employed a platform that encodes quantum information within the mechanical oscillator motion of a single trapped ion, in effect optimizing the generation and control of logical states of Gottesman–Kitaev–Preskill code for finite-energy states. The approach supported efficient correction of unwanted displacements in the oscillator's motion, and lengthened coherence time threefold.
|
Eye Provides Clues to Vascular Disease
University of Bonn (Germany)
February 11, 2022
Atherosclerosis (hardening of the arteries) could be diagnosed through the use of self-learning software devised by researchers at Germany's University of Bonn and University Hospital of Bonn to identify vascular changes in images of the eye. The researchers fed training images of the eyes of patients with early-stage peripheral arterial disease (PAD) to a convolutional neural network (CNN). The University of Bonn's Thomas Schultz said the software correctly identified "a good 80% of all affected individuals, if we took into account 20% false positives." He added that this is impressive "because even for trained ophthalmologists, PAD can't be detected from fundus images." Further analysis showed the CNN focuses especially on large vessels in the back of the eye, although digital images needed to be of sufficiently high resolution to realize optimal results.
|
Sony AI Drives Race Car Like a Champ
Wired
Will Knight
February 9, 2022
Researchers at Sony and Polyphony Digital say their GT Sophy is the first artificial intelligence (AI) able to beat professional esports players in the motorsport game Gran Turismo. While AI has mastered board games like chess or Go, mastering video games like Gran Turismo requires high-speed reflexes and the ability to make continuous judgments. Stanford University's Chris Gerdes said, "Outracing human drivers so skillfully in a head-to-head competition represents a landmark achievement for AI." Gerdes said the techniques used to develop GT Sophy, which mastered Gran Turismo through hours of practice, could be applied to the development of autonomous cars. However, University of Massachusetts Amherst's Bruno Castro da Silva said it is difficult to guarantee the safety and reliability of reinforcement learning algorithms like GT Sophy in the real world. Said da Silva, "A lack of safety guarantees is one of the main reasons why machine learning-based robots are not yet widely used in factories and warehouses."
|
|
