Welcome to the June 2, 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."
|
|
The Race to Make AI Smaller, Smarter
The New York Times Oliver Whang May 30, 2023
The BabyLM Challenge, organized by computer scientists at institutions including Johns Hopkins University and Switzerland's ETH Zurich, is aimed at creating more accessible, intuitive language models, in stark contrast to the race for ever-larger language models undertaken by big tech companies. The goal is to produce a mini-language model using datasets less than one-ten-thousandth the size used by most advanced large language models. As part of the challenge, researchers have been tasked with training language models on about 100 million words, with the winning model to be chosen based on the effectiveness of their generation and understanding of the nuances of language.
*May Require Paid Registration
|
Shared Modeling Can Help Schools Predict, Avert Dropouts
Cornell Chronicle Tom Fleischman June 1, 2023
A multi-institutional team of researchers found under-resourced schools can anticipate and prevent student dropouts through a shared educational modeling framework. Cornell University's Rene Kizilcec and colleagues converted anonymized data from four U.S. universities into a common structure to predict which students are most likely to drop out, sharing only university-specific models between researchers. The team evaluated the success of transfer learning using direct, voting, and stacked transfer approaches, as well as local modeling at each of the four universities. Kizilcec said local modeling better predicted dropout rates at a lower-than-expected level given the institutions' differing sizes, graduation rates, and student demographics. Kizilcec said the outcomes indicate greater equity in dropout forecasting, which could help under-resourced institutions with earlier intervention and avert dropouts.
|
The Ranking of Computer Science Rankings
Forbes Matt Symonds May 31, 2023
U.K.-based knowledge resource BlueSky Thinking compiled the four major global subject rankings published by Times Higher Education, Quacquarelli Symonds, U.S. News, and China's Academic Ranking of World Universities into its Ranking of Computer Science (CS) Rankings 2022/23. The combined ranking has U.S. universities in its top four spots, with Harvard, Princeton, and Cornell universities, as well as the University of Washington, all in the top 15. All four rankings consistently rate the Massachusetts Institute of Technology, Stanford University, and Carnegie Mellon University at the top, while European, Canadian, and Asia-Pacific institutions comprise nearly 60% of the world's top 50 CS rankings.
|
Supercomputer Simulations Provide Better Picture of Sun's Magnetic Field
Aalto University (Finland) June 1, 2023
Researchers at Finland's Aalto University and Germany's Max Planck Institute for Solar System Research (MPS) used petascale supercomputers in those countries to directly simulate a small-scale dynamo in the Sun. Understanding how magnetic fields are generated and amplified inside the Sun could help improve predictions of major solar events. Prior studies indicated the small-scale dynamo may not be able to exist in the Sun due to its very low magnetic Prandtl number (PrM), but the researchers found a small-scale dynamo is possible by simulating conditions of turbulence with unprecedentedly low PrM values. MPS's Jörn Warnecke said, "This result will bring us closer to resolving the riddle of [coronal mass ejection] formation, which is important for devising protection for the Earth against hazardous space weather."
|
Making Delivery Robots Disability-Friendly
BBC News Beth Rose May 30, 2023
Estonian robotic delivery company Starship Technologies has deployed disability-friendly robots to deliver purchases in the U.K. city of Wakefield. Starship's Lisa Scott said the company has implemented safety protocols and programmed the robots to be "cautious pedestrians" by using sensors and cameras to track obstacles. Scott noted the company "spent a lot of time having the robots learn what mobility devices look like" so they know to get clear of them. Starship hopes the robots will become more adept at recognizing the canes used by visually impaired people through on-the-job training. The company established a Disability Advisory Panel that includes disabled members to ensure all stakeholders have a voice in the technology's development and improvement.
|
Willamette University Creates School of Computing, Information Sciences
Portland Business Journal Malia Spencer May 23, 2023
Willamette University’s new School of Computing and Information Sciences will house its undergraduate program in computing and data science, as well as its master's program in data science. The institution also is adding a computer science (CS) master's degree and is developing new undergrad and graduate programs in CS and statistics. Wrote Willamette president Stephen E. Thorsett, "The formal creation of the new School helps us enhance our existing offerings [in CS and information science] in a way that not only helps meet the critical demands of the workforce but equips our students—from all backgrounds and career pathways—with the technological skills they'll need to fulfill their mission of building a better world." Thorsett said the school will help students gain not just coding skills, but also ethical acumen and a human-centered approach to their work.
|
Training Machines for Uncertain Real-World Situations
MIT News Adam Zewe May 31, 2023
An algorithm developed by researchers at the Massachusetts Institute of Technology (MIT) and Technion—Israel Institute of Technology can determine automatically and independently whether and when imitation learning or reinforcement learning is more effective for training a "student" machine. The algorithm is adaptive, allowing the machine to move between both types of learning throughout the training process based on which would achieve better, faster results. MIT's Idan Shenfeld said, "This combination of learning by trial-and-error and following a teacher is very powerful. It gives our algorithm the ability to solve very difficult tasks that cannot be solved by using either technique individually."
|
AI Model Digests Video to Learn Sign Language
New Atlas Paul McClure May 24, 2023
A tool developed by researchers at Spain’s Barcelona Supercomputing Center and the Universitat Politècnica de Catalunya (UPC) uses artificial intelligence (AI) to improve sign language translation. The researchers fed a transformer-style machine learning model more than 80 hours of video in American Sign Language with corresponding English transcripts from the How2Sign dataset. The researchers addressed the variability and complexity of sign languages by pre-processing the video data to extract spatiotemporal information via the Inflated 3D Networks technique. They found the model produced meaningful translations, although they agreed, "There is still room for improvement."
|
Low-Cost Human Biomarker Sensors Designs
Penn State News Ashley WennersHerron May 31, 2023
Pennsylvania State University (Penn State) engineers have designed RNA-based human biomarker sensors that, for about US$1, can conduct measurements that conventional tests would cost anywhere from $100 to more than $1,000 to execute. The technology combines a cell-free expression system with riboswitches, at a cumulative cost of about $1 per reaction. Penn State's Howard Salis said, "Using a combination of thermodynamic modeling and computational optimization, we rationally designed new riboswitches that are predicted to be excellent protein sensors and then we tested them." Results showed the sensors can detect human biomarker proteins, including the MS2 protein contained in a bacterial phage, monomeric C-reactive protein involved in chronic inflammatory conditions, and interleukin-32 gamma that indicates viral or bacterial infection.
|
The Making of a Mona Lisa Hologram
AIP Publishing May 30, 2023
An international team of researchers developed an acoustic metasurface-based holography technique, which they used to recreate the Mona Lisa. The researchers used a deep neural network-based algorithm to customize the metasurface's antenna-like components. Said Yue-Sheng Wang of China's Tianjin University, "A metasurface-based hologram works by precisely controlling the phase and amplitude of the waves interacting with the metasurface. As a result, the outgoing waves at each pixel exhibit a certain amplitude and phase, which results in the desired holographic image based on their interference." As a proof of concept, the researchers used the technique to reconstruct the Mona Lisa, and to recreate her left eye in great detail.
|
Sweater-Wrapped Robots Can Feel, React to Human Touch
Carnegie Mellon University School of Computer Science Stacey Federoff May 25, 2023
The RobotSweater machine-knitted textile "skin" fabricated by Carnegie Mellon University (CMU) researchers can sense contact and pressure "to make the robot smarter during its interaction with humans," according to CMU's Changliu Liu. The fabric includes two layers of conductive yarn containing metallic fibers, with a lace-patterned layer sandwiched between them. Applying pressure to the fabric causes the yarn to close a circuit, which is read by sensors. The fabric can be tailored to conform to uneven three-dimensional surfaces. Liu said the material can cover the robot's entire body so it can sense any possible collisions, which can be especially useful in environments where safety is prioritized.
|
Microsoft Finds macOS Bug That Lets Hackers Bypass SIP Root Restrictions
BleepingComputer Sergiu Gatlan May 30, 2023
Apple has patched a vulnerability discovered by Microsoft security researchers, dubbed Migraine, that would have allowed attackers with root privileges to install "undeletable" malware and access the victim's private data. The researchers said, "By focusing on system processes that are signed by Apple and have the com.apple.rootless.install.heritable entitlement, we found two child processes that could be tampered with to gain arbitrary code execution in a security context that bypasses SIP [System Integrity Protection] checks." Bypassing SIP also would allow attackers to circumvent Transparency, Consent, and Control (TCC) policies to gain access the victim's private data. The vulnerability was patched in Apple's May 18 security updates for macOS Ventura 13.4, macOS Monterey 12.6.6, and macOS Big Sur 11.7.7.
|
Wireless Ultrasound Monitoring for Subjects in Motion
UC San Diego Today Emerson Dameron May 22, 2023
Engineers at the University of California, San Diego and the Massachusetts Institute of Technology have developed a fully integrated ultrasound sensor that can conduct deep-tissue monitoring of subjects in motion. The autonomous, wearable ultrasonic system-on-patch (USoP) features a flexible control circuit that interacts with an ultrasound transducer array to gather and transmit data wirelessly, while a machine learning algorithm interprets the data and monitors moving subjects. The algorithm can analyze the received signals and select the most suitable channel to keep track of the subject. The researchers found the USoP enables continuous tracking of physiological signals like central blood pressure, heart rate, and cardiac output from tissues as deep as 164 millimeters, for up to 12 hours at a time.
|
|