Welcome to the September 21, 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."
|
|
We Can Train Big Neural Networks on Small Devices
IEEE Spectrum Matthew Hutson September 20, 2022
A new training method expands small devices' capabilities to train large neural networks, while potentially helping to protect privacy. The University of California, Berkeley's Shishir Patil and colleagues integrated offloading and rematerialization techniques using suboptimal heuristics to reduce memory requirements for training via the private optimal energy training (POET) system. Users feed POET a device's technical details and data on the architecture of a neural network they want to train, specifying memory and time budgets; the system generates a training process that minimizes energy usage. Defining the problem as a mixed integer linear programming challenge was critical to POET's effectiveness. Testing showed the system could slash memory usage by about 80% without significantly increasing energy consumption.
|
A New Method for Denoising Images
Gwangju Institute of Science and Technology (South Korea) September 13, 2022
Researchers from South Korea's Gwangju Institute of Science and Technology (GIST), Vietnam's VinAI Research, and Canada's University of Waterloo have developed a reference-free image denoising method. The new approach uses a post-correction network and a self-supervised machine learning framework to improve the quality of path-traced visuals. The model tripled the quality of rendered images relative to input images by preserving finer details, and could be trained on the fly to output final images in just 12 seconds. Said GIST's Bochang Moon, "Our approach is the first that does not rely on pre-training with an external dataset. This, in effect, will shorten the production time and improve the quality of offline rendering-based content such as animation and movies."
|
YouTube's Dislike Button Rarely Shifts Recommendations
The New York Times Nico Grant September 20, 2022
Mozilla Foundation researchers found pressing the "dislike" button on a YouTube video does little to improve the videos chosen by the YouTube platform's recommender algorithm. The researchers analyzed over 567 million YouTube video recommendations, and used Mozilla's RegretReporter tool to examine the recommendation algorithm. They found use of the dislike button reduced similar, unwanted recommendations by just 12%. Selecting "Don't recommend channel" was 43% effective in reducing unwanted recommendations, while pressing "not interested" was shown to be 11% effective, and removing videos from the user’s watch history was 29% effective. Mozilla's Jesse McCrosky said YouTube should incorporate greater transparency, and should provide users with more control over what they see.
*May Require Paid Registration
|
Fairer Ranking System Diversifies Search Results
Cornell Chronicle Patricia Waldron September 19, 2022
Cornell University researchers Yuta Saito and Thorsten Joachims have developed a fairer ranking system for recommendations that prevents search results from only highlighting a few top hits. Conventional recommender systems try ranking items based on what users want to see, so Saito designed the enhanced ranking system according to the economic principles of "fair division." The researchers used synthetic and real-world data to test the system's feasibility. The system returns viable results that rank items' benefits better than random discovery, does not easily improve items' impact, and does not confer an advantage to any item by switching how it is ranked versus other items in a series of searches. Saito said the framework "can be applied to any type of two-sided ranking system."
|
China's Factories Accelerate Robotics Push as Workforce Shrinks
The Wall Street Journal Jason Douglas September 18, 2022
A report from the International Federation of Robotics (IFR) revealed more than 243,000 industrial robots were shipped to China last year, up 45% from 2020. China, the top market worldwide for robot manufacturers, accounted for slightly less than half of all heavy-duty industrial-robot installations in 2021, which was nearly double the number installed in the Americas and Europe last year combined. However, the U.S., Japan, Germany, and South Korea still have more robots on production lines than China. The rapid growth of automation in China can be attributed to a rush to catch up to those countries, as well as adapting to a decline in its workforce as its population shrinks and more younger workers opt for service jobs over manufacturing.
*May Require Paid Registration
|
ML Generates 3D Model from 2D Pictures
Washington University in St. Louis Brandie Jefferson September 19, 2022
A machine learning algorithm developed by researchers at Washington University in St. Louis can use a partial set of two-dimensional (2D) images produced by standard microscopy tools to create a continuous three-dimensional (3D) model of cells. The researchers accomplished this through the use of a neural field network, which can represent a sample in its entirety using a limited number of 2D images, eliminating the need to save data-heavy image files. The imaging system also has the ability to zoom in on a pixelated image and recreate missing pieces. Washington University's Ulugbek Kamilov said, "We train the model on the set of digital images to obtain a continuous representation. Now, I can show it any way I want. I can zoom in smoothly and there is no pixelation."
|
Deepfake Audio Has a Tell
Ars Technica Logan Blue; Patrick Traynor September 20, 2022
Researchers at the University of Florida can detect audio deepfakes by measuring acoustic and fluid dynamic distinctions between organic and synthetic voice samples. The researchers inverted techniques used to replicate the sounds a person makes to acoustically model their vocal tract, in order to approximate the speaker's tract during a segment of speech. Using the process to analyze deepfaked audio samples, on the other hand, can result in model vocal tract shapes that do not appear in people. "By estimating the anatomy responsible for creating the observed speech, it's possible to identify whether the audio was generated by a person or a computer," the researchers explain.
|
Test Uses Nanotechnology, AI to Diagnose TB in Children
Tulane News September 16, 2022
An international team led by researchers at Tulane University leveraged nanotechnology and artificial intelligence (AI) to create a smartphone-based blood test to diagnose tuberculosis (TB) in children at the point of care. The researchers coated nanoparticles with antibodies that bind components of the bacteria that cause TB (the lipoarabinomannan molecule and the associated protein LprG), which can be seen under a microscope. They also developed an AI algorithm to remove background noise caused by other substances on nanoparticle surfaces and analyze samples quickly. The test detected TB in 89% of children known to have the disease, and identified 74% of children whose TB was missed by standard tests.
|
Danish Initiative Aims to Develop First Quantum Computer for Life Sciences
Financial Times Clive Cookson September 21, 2022
Denmark's Novo Nordisk Foundation has committed to investing $200 million to develop what it describes as the first practical quantum computer for life sciences research. The nonprofit says the seven-year initiative will assess competing technologies before choosing which one to use. Said Peter Krogstrup, who will lead the program from the Niels Bohr Institute at Denmark's University of Copenhagen, "We would rather spend seven years finding the platform that offers the greatest opportunity to build a usable quantum computer." Scientists at Danish institutions and universities in other countries, including the Netherlands' Technical University of Delft and Canada's University of Toronto, will participate in the program.
*May Require Paid Registration
|
Engineers 3D-Print Highest Specific-Strength Titanium Alloy
Monash University (Australia) September 19, 2022
Engineers at Australia's Monash University three-dimensionally (3D)-printed a titanium alloy with the highest strength-to-weight ratio of all 3D-printed metals to date. The researchers used additive manufacturing to manipulate the unique microstructures of commercial titanium alloys. Monash's Aijun Huang said post-heat treatment of one commercial alloy yielded elongation and tensile strengths of more than 1,600 megapascals (MPa). "Our findings offer a completely new approach to precipitation strengthening in commercial alloys that can be utilized to produce real components with complex shape for load-bearing application," Huang explained. "This application is still absent for any titanium alloys to date."
|
Face Recognition for Pigs Could Improve Welfare on Farms
New Scientist Christa Lesté-Lasserre September 20, 2022
Wenhao Zhang and colleagues at the U.K.'s University of the West of England have developed a facial recognition system for pigs. The researchers captured thousands of photos of sows from three breeds, then fed the images into a convolutional neural network that detected nuanced distinctions in the animals' faces. The software ran tens of millions of tests in learning to identify individual pigs, resulting in a system that, Zhang said, could accurately identify specific animals nearly 97% of the time. Natalie Waran at New Zealand's Eastern Institute of Technology said facial recognition makes physically marking the animals unnecessary, while also offering the opportunity to "cater to an individual's specific [nutritional or veterinary] needs where animals are kept in large groups in intensive farming conditions."
|
Software Platform Advances Understanding of Manufactured Components' Surface Finishes
University of Pittsburgh Swanson Engineering September 19, 2022
Scientists from Germany's University of Freiburg and the University of Pittsburgh have standardized surface analysis with the new contact.engineering software platform. The platform allows users to generate a digital twin of an engineered surface in order to predict its properties. It integrates data from different tools, fixes measurement errors, and uses the data submitted to build the digital twin. The platform calculates statistical metrics and applies mechanical models to the surfaces, which Freiburg's Lars Pastewka said enables users to "identify which topographical features influence which properties. This allows a systematic optimization of finishing processes." Users also can upload, analyze, and share measurements via the platform.
|
Implantable Sensor Sniffs Out Possible Signals of Osteoarthritis
Pennsylvania State University Ashley J. WennersHerron September 14, 2022
A flexible biosensor was developed by researchers at Pennsylvania State University (Penn State) and several universities in China to perform continuous, wireless monitoring of nitric oxide, a biomarker associated with inflammation, in rabbits. The device is comprised of flexible and biocompatible organic electrochemical transistors and a channel made of the polymer PEDOT:PSS. The sensors can transmit nitric oxide levels to a smartphone app via Bluetooth. Penn State's Huanyu "Larry" Cheng said, "The reference-free sensor with a miniaturized active sensing region enables nitric oxide detection with improved spatial resolution compared with previously reported electrochemical nitric oxide sensors, which could allow the mapping of electrochemical signals to offer comprehensive diagnostic information."
|
|