Welcome to the July 31, 2023, 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."
|
|
|
Astronaut Controls Earth Robots While Aboard ISS
Interesting Engineering
Chris Young
July 27, 2023
The European Space Agency (ESA) blogged that U.S. National Aeronautics and Space Administration (NASA) astronaut Frank Rubio controlled a group of earthbound robots while aboard the International Space Station. During the two-hour test, Rubio teleoperated three robots in a simulated lunar environment at the German Aerospace Center (DLR) in Munich. The ESA said he utilized "a force-feedback device" to remove a seismometer from a lunar lander and place it on the ground using DLR's Rollin' Justin humanoid robot and ESA's four-wheeled, two-armed Interact rover. The tasks were part of a series of tests in preparation for a follow-up mission that will utilize a larger team of robots.
|
Fiber-Infused Ink Enables 3D-Printed Heart Muscle to Beat
Harvard University John A. Paulson School of Engineering and Applied Sciences
Kat J. McAlpine
July 27, 2023
Scientists at Harvard University, Boston Children's Hospital, and Germany's Technical University Munich developed a fiber-infused gel (FIG) ink that can be used to three-dimensionally (3D)-print a ventricle able to beat like a human heart. Harvard's Suji Choi said the ink supports such complex configurations without additional materials or scaffolds because it maintains its 3D shape once printed. Choi used a rotary jet spinning method to produce a sheet of material whose appearance resembled that of cotton, then sonically broke it into tiny fibers dispersed into the FIG ink. Choi learned printing two-dimensional and 3D structures with the ink aligned the heart muscle cells with the direction of the ink's fibers, and electrically stimulating the 3D structures induced contractions.
|
Llama, ChatGPT Are Not Open-Source
IEEE Spectrum
Michael Nolan
July 27, 2023
Researchers at the Netherlands' Radboud University assessed 21 large language models (LLMs) ostensibly designated as open source, and learned most models' openness is more limited than purported. The researchers found OpenAI's ChatGPT scored worst for openness, labelling it "closed" in all evaluations except for model card and preprint, which received "partial" status. Meta's Llama 2 is the second-worst-scoring LLM, despite the social media company's claim the release aimed to make the model "accessible to individuals, creators, researchers, and businesses so they can experiment, innovate, and scale their ideas responsibly." Although several smaller, research-oriented models were found to be more open than ChatGPT or Llama 2, the researchers said few disclosed reinforcement learning with human feedback functions in sufficient detail, and most models were not peer reviewed.
|
Ringer Awarded SIGPLAN Distinguished Service Award
University of Illinois Grainger College of Engineering Computer Science
Bruce Adams
July 25, 2023
The University of Illinois at Urbana-Champaign's Talia Ringer received the Distinguished Service Award from the ACM Special Interest Group on Programming Languages (SIGPLAN) for contributions that included founding and chairing the SIGPLAN Long-Term Mentoring Committee program. The international program matches coding language mentors with mentees from different institutions for engagement that lasts at least 12 months. ACM SIGPLAN also cited the Computing Connections Fellowship program that the committee launched. Google DeepMind's Jeff Dean lauded Ringer for her "incredible dedication to many different research communities, to mentoring new researchers, and to bringing people together."
|
Software Designs Optimize, Personalize Treatments for Movement Impairments
Rice University News
Silvia Cernea Clark
July 27, 2023
Physicians can use open source software released by Rice University engineers to optimize and personalize treatments for neurologic and orthopedic mobility impairments by simulating how individual patients move. The Neuromusculoskeletal Modeling Pipeline software incorporates models of central nervous system control, muscle force generation, and metabolic energy expenditure. The researchers made multiple predictions about how a stroke patient would walk at their fastest-comfortable speed of 0.8 meters (2.6 feet) per second based on post-stroke walking at their self-selected speed of 0.5 meters (1.6 feet) per second; they formulated the best course of treatment from these predictions. Rice's B.J. Fregly said, "Clinicians working with engineers could use our software to make explicit, objective predictions, which could not only weed out ineffective or harmful treatments but also generate highly effective unforeseen ones."
|
Complex-Domain Neural Network Advances Large-Scale Coherent Imaging
SPIE
July 27, 2023
A team of researchers from China's Beijing Institute of Technology, the California Institute of Technology, and the University of Connecticut has augmented large-scale coherent imaging with a complex-domain neural network. The researchers integrated a two-dimensional complex convolution unit and complex activation function into a network that generates multidimensional representations of complex wavefront. They also created a multi-source noise model that advances domain-adaptation ability from synthetic to real data. The researchers found the approach reduces exposure time and data volume by an order of magnitude without hindering efficiency or reconstruction quality.
|
Researchers Leverage AI to Fight Online Hate Speech
University of Michigan Computer Science and Engineering
July 28, 2023
Researchers at the University of Michigan (U-M) worked with colleagues at Microsoft to create a tool for detecting online hate speech by combining deep learning models and traditional rules-based strategies. The Rule By Example (RBE) approach "pairs logical rules that are very explainable with [hate speech exemplars] and then encodes and learns them," said U-M's Christopher Clarke. RBE can accurately forecast and categorize online hate speech using rule and text encoders to learn sound, accurate implantations of hateful content and their underlying rules. The framework also builds-in transparency by allowing users to view the factors shaping the model's precision. RBE was 2% more accurate than the closest rival classifier.
|
Quantum Tech Will Transform National Security. It’s Testing U.S. Alliances Now
The New York Times
Damien Cave
July 29, 2023
The race to develop quantum technologies is accelerating as governments boost investments and scientists make rapid advances but is threatened by protectionist policies. The U.S., for example, is considering tougher export controls for quantum technologies, focusing particularly on China but impacting allies as well. Australia, meanwhile, has been exploring how to keep its own advances secret. Critics of such policies say further limits, on top of those already in place, could stifle momentum and even backfire because the strength of the U.S. model of tech development comes from its openness, combining pools of public research money with private investment to support scientists around the globe. Said Michael Biercuk, the founder of Q-CTRL, a start-up that recently signed a deal with Australia’s Department of Defense to develop and field-test its quantum sensor technology, "Anytime we overly regulate emerging areas of science, we risk simply stopping progress locally and ceding technological advantage to our adversaries."
|
Researchers Use Motion to Discover Objects in Videos
Carnegie Mellon University School of Computer Science
Byron Spice
July 26, 2023
Researchers at Carnegie Mellon University demonstrated computer vision systems' ability to discover moving objects more easily in videos than stationary objects with the new MoTok framework. The researchers developed MoTok with colleagues at the Toyota Research Institute, the University of California, Berkeley, and the University of Illinois in Urbana-Champaign to identify features of objects it perceives as moving. The framework then rebuilds the object based on these features, enabling the computer to rediscover it. The researchers also enabled MoTok to simplify object virtualization to better identify high-level properties and categorize objects.
|
Android Malware Steals User Credentials Using Optical Character Recognition
Ars Technica
Dan Goodin
July 28, 2023
Malicious Android applications uncovered by researchers at security firm Trend Micro steal user credentials on phone screens via optical character recognition (OCR). The researchers found at least four Android apps available outside of Google Play hosting the CherryBlos malware, which used a paid version of Jiagubao software to encrypt code and code strings to block detection. CherryBlos overlaid windows that emulated those of legitimate apps for Binance and other cryptocurrency services, and substituted the wallet addresses the victim chose to receive funds with attacker-controlled addresses. The researchers also discovered the malware can record mnemonic passphrases for accessing accounts, using OCR to render images as text. They also found 21 Google Play-hosted apps using the same digital certificate or attacker framework as the CherryBlos apps.
|
Algorithm Maps Safest Routes for City Drivers
University of British Columbia (Canada)
July 25, 2023
A new algorithm developed by researchers at Canada's University of British Columbia (UBC) utilizes real-time crash risk data to identify the safest route in an urban network. The researchers detected near-misses between vehicles in downtown Athens, Greece, then forecast collision risks using data collected from 10 hovering drones. UBC's Tarek Sayed said, "The algorithm is capable of adjusting directions in real time, suggesting detours to avoid hazardous locations. This helps enhance road safety for all users." The researchers found the safest routes are not necessarily the quickest; in one case, analysis of a small segment of Athens' urban road network determined just 23% of the fastest routes were also deemed the safest.
|
Resource Harmonizes 16S, Shotgun Sequencing Data for Microbiome Research
UC San Diego Today
Nicole Mlynaryk
July 27, 2023
University of California, San Diego (UCSD) scientists unveiled the Greengenes2 reference database, which makes microbiome data comparable and integrative by correlating the 16S ribosomal RNA gene amplicon and shotgun metagenomics sequencing techniques. Said UCSD's Rob McKnight, "Standardizing results across these two methods will significantly improve our chances of discovering microbiome biomarkers for health and disease." The researchers developed Greengenes2 with U.S. and international colleagues, expanding the Web of Life whole genome database before applying new computational tools to integrate existing full-length 16S sequences into the whole-genome phylogeny. They entrenched 16S fragments from over 300,000 microbiome samples with another machine learning tool into a database for organizing 16S and shotgun sequencing data.
|
|
