Welcome to the March 13, 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."
|
|
|
'Traffic Cop' Algorithm Helps Drone Swarm Stay on Task
MIT News
Jennifer Chu
March 13, 2023
The WiSwarm scheduling algorithm developed by Massachusetts Institute of Technology and Columbia University engineers can tailor any wireless network to manage a high volume of time-sensitive data from multiple sources. The researchers aimed to prioritize age-of information via a "last in, first out" protocol for multiple robots collaborating on time-sensitive tasks over conventional wireless networks. WiSwarm can run on a centralized computer and be coupled to any wireless network to govern multiple data streams, and to prioritize the freshest data. The algorithm prioritizes data by evaluating and multiplying a drone's priority, age of information, and probability of successfully sending data at any given time.
|
AI Re-Creates What People See by Reading Brain Scans
Science
Kamal Nahas
March 7, 2023
The Stable Diffusion artificial intelligence (AI) algorithm developed by German and Japanese researchers can read functional magnetic resonance imaging (fMRI) brain scans to replicate images people have seen recently. Yu Takagi at Japan's Osaka University said the algorithm employs information collected from brain regions involved in image perception as the fMRI scan records peaks in brain activity; AI then is used to translate these patterns into an imitation image. The researchers further trained Stable Diffusion on a University of Minnesota dataset of four people viewing a series of 10,000 photos. To address the algorithm's tendency to render objects in photos as abstract figures, the researchers fed keywords from image captions accompanying the photos to the text-to-image generator.
|
Regulators Look at Potential Bias in Life Insurance Algorithms
The Wall Street Journal
Leslie Scism
March 12, 2023
State regulators are taking issue with U.S. life insurers' use of data science to expedite applications, concerned that the companies' algorithms could be biased against minorities. Many insurers are algorithmically underwriting life policies to assess applicants' health quickly, and to improve the convenience of sales for consumers who favor shopping online. Executives say insurers usually use the same data as human underwriters in algorithmic-based systems, but regulators are worried about a wealth of unconventional information hyped by data vendors. Colorado's Division of Insurance has started developing rules under a new statute limiting insurers' use of algorithms, predictive models, and the data they amass. Last year, Connecticut started requiring insurers to certify their data use fully adheres to antidiscrimination laws.
*May Require Paid Registration
|
Researchers Use Table Tennis to Understand Human-Robot Dynamics in Agile Environments
Georgia Institute of Technology
Breon Martin
March 8, 2023
Researchers at the Georgia Institute of Technology (Georgia Tech) developed a collaborative robot (cobot) that uses table tennis to demonstrate that robots and humans can collaborate on tasks. The Barrett WAM robotic arm, equipped with a camera and capable of holding a paddle, was trained using imitation learning, with positive reinforcement given for successful volleys and negative reinforcement for unsuccessful volleys. Georgia Tech's Matthew Gombolay said, "We leveraged prior work on table tennis and 'learning from demonstration techniques' in which a human can teach a robot a skill, such as how to hit a table tennis shot or simply having the human demonstrate the task to the robot."
|
Courses in the Metaverse Struggle to Compete with Real World
Financial Times
Jonathan Moules
March 12, 2023
Business schools offering courses in the metaverse and other virtual technologies must compete with real-world study. For example, in-person learning works better for teaching interpersonal skills vital to postgraduate courses, like networking and leadership; the metaverse is currently snarled in widespread rejection for failing to live up to hype. Many business schools are testing out metaverse concepts, in some cases partnering with gaming and virtual reality technology companies to develop teaching platforms. A global alliance of business schools called the Future of Management Education includes Belgium's Vlerick Business School, which teamed with a local gaming firm to build a virtual teaching environment for master’s degree students. Students at the U.K.'s Imperial College Business School, meanwhile, are testing avatar role-play with kits from London-based startup Bodyswaps.
*May Require Paid Registration
|
Open Source Software Could Deliver Huge Time Savings for Computational Chemists
Chemistry World
Fernando Gomollón-Bel
March 7, 2023
Software developed by a team led by researchers at Spain's Institute of Chemical Synthesis and Homogeneous Catalysis (ISQCH) aims to make computational chemistry experiments more efficient and speed up automated analyses. The open source software, AQME, requires only simple inputs, like a Smiles string, which converts complex chemical structures into machine readable successions of letters and numbers. This could enable integration with chemical databases and machine learning solutions, as many include datasets in the Smiles format. ISQCH's Juan Vicente Alegre Requena said, "Most of the outputs are graphs and plots, which provides an immediate visualization of the results. AQME also automatically analyzes the existence of errors, corrects the calculation files, and re-launches the simulations."
|
Accelerating Data Retrieval in Huge Online Databases
MIT News
Adam Zewe
March 13, 2023
A team of researchers from the Massachusetts Institute of Technology, Harvard University, and Germany's Technical University of Munich designed machine learning (ML) hash functions that can accelerate online database searches. The researchers found using learned models rather than traditional hash functions could halve the collisions between data items with identical hash value and provide greater computational efficiency than perfect hash functions. They used ML to approximate the distribution of a small sample taken from a dataset, which the learned model employs to predict the location of a key in the dataset. Learned models could shrink the ratio of colliding keys in a dataset from 30% to 15% versus traditional hash functions when data was predictably distributed. They also trimmed nearly 30% off the runtime in the best cases.
|
Google's PaLM-E Generalist Robot Brain Takes Commands
Ars Technica
Benj Edwards
March 7, 2023
Researchers at Google and Germany's Technical University of Berlin debuted PaLM-E, described as the largest visual-language model (VLM) ever created. The multimodal embodied VLM contains 562 billion parameters and combines vision and language for robotic control; Google claimed it can formulate a plan of action to execute high-level commands using a mobile robot platform equipped with an arm. PaLM-E analyzes data from the robot's camera without requiring pre-processed scene representations, eliminating human data pre-processing or annotation. The VLM's integration into the control loop also instills resistance to interruptions during tasks. PaLM-E encodes continuous observations into a sequence of vectors identical in size to language tokens, so it can "understand" sensor data in the same way it processes language.
|
High-Speed Super-Resolution Microscopy via Temporal Compression
SPIE Newsroom
March 10, 2023
A multi-institutional team of Chinese scientists has resolved the disconnect between spatial resolution and imaging speed in optical microscopy through temporal compressive super-resolution microscopy (TCSRM). The method combines augmented temporal compressive microscopy with deep-learning-based super-resolution image reconstruction. The first process enhances imaging speed by reassembling multiple images from one compressed image, while the second process supports super-resolution without slowing imaging speed. The researchers' iterative image reconstruction algorithm performs motion estimation, merging estimation, scene correction, and super-resolution processing to generate the super-resolution image sequence from compressed and reference quantifications. The team used TCSRM to image flowing fluorescent beads in a microchannel, realizing a 1,200-frames-per-second rate and 100-nanometer spatial resolution.
|
Transistor Could Shrink Communications Devices on Smartphones
University of Michigan News
Catherine June
March 8, 2023
University of Michigan (U-M) scientists have engineered a reconfigurable transistor that incorporates a ferroelectric semiconductor with the nanoscale thinness required for modern computing elements. Ferroelectric semiconductors can maintain electrical polarization while also switching positive and negative ends, enabling the transistor to adjust its behavior. U-M's Ding Wang said the multifunctional ferroelectric high electron mobility transistor (FeHEMT) can reduce circuit area, cost, and power consumption. U-M's Zetian Mi said the FeHEMT has the potential "for integrating multifunctional devices, such as reconfigurable transistors, filters, and resonators, on the same platform—all while operating at very high frequency and high power."
|
|
