Welcome to the July 12, 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."
|
|
|
ACM Issues Principles for Generative AI Technologies
ACM
July 11, 2023
ACM's global Technology Policy Council (TPC) released "Principles for the Development, Deployment, and Use of Generative AI Technologies" in response to innovations in generative artificial intelligence (AI) and their ramifications. The statement lists eight principles for cultivating fair, accurate, and advantageous decision-making for generative and all other AI. Principles applicable to generative AI include imposing limits and guidance on deployment and use; accounting for the technology's structure and function in intellectual property law and regulation; personal control of data; and correctability of errors via public repositories provided by generative AI system developers. TPC's Ravi Jain added, "We must also build a community of scientists, policymakers, and industry leaders who will work together in the public interest to understand the limits and risks of generative AI as well as its benefits."
|
NY Fed Says Test on Digital Dollar Shows Speed Advantage
Bloomberg
Jennifer Surane; Yueqi Yang
July 6, 2023
The Federal Reserve Bank of New York's Innovation Center said a 12-week test of a regulated liability network proved the use of digital dollars could enhance domestic and cross-border transactions. Banks could simulate issuing digital tender representing customers' own money with the network before going through central bank reserves on a distributed ledger. Major banks like Citigroup and Wells Fargo participated in the test, which the researchers conducted on a private blockchain accessed by permission. Participants also learned the network could help synchronize dollar-denominated payments and enable near-real-time settlements. The project reportedly did not find "any insuperable legal impediments" to establishing a digital dollar under current U.S. statutes.
*May Require Paid Registration
|
Natural World Empowers Future of Computer Vision
Princeton University
Adam Hadhazy
July 6, 2023
An open source software system developed by Princeton University researchers aims to improve the training of computer vision systems by quickly generating an unlimited number of photorealistic scenes of the natural world. The Infinigen system uses randomized mathematical rules to produce natural-looking three-dimensional (3D) objects and synthetic representations of natural phenomena. Each image is generated programmatically, which involves developing a virtual world and populating it with digital objects, which allows for automatic labeling and training computer vision systems to identify and locate objects with only an image as input. Princeton's Jia Deng said the researchers expect Infinigen will prove “a useful resource not just for creating training data for computer vision, but also for augmented and virtual reality, game development, film-making, 3D printing, and content generation in general."
|
Green-Screen Filming Method Uses Magenta Light
New Scientist
Matthew Sparkes
July 7, 2023
Researchers at streaming service Netflix have developed a way to instantly generate visual effects for film and TV using a new green-screen technology powered by artificial intelligence (AI). The Magenta Green Screen technique involves filming actors with bright green light-emitting diodes (LEDs) from the back and red and blue LEDs from the front, producing a magenta glow. Film editors can replace the background-recording green channel in real time, splicing the actors into the foreground of another scene without difficulty, even with potentially problematic areas. Netflix uses AI to remove the actors' magenta tint by restoring a normal range of color to the foreground, utilizing a photo of the actors under normal illumination as reference for a realistic-appearing green channel.
|
Training Robots to Make Decisions on the Fly
University of Illinois Urbana-Champaign
Debra Levey Larson
July 6, 2023
A learning-based method developed by researchers at the University of Illinois Urbana-Champaign (UIUC) allows autonomous landers to decide where and how to land and collect terrain samples based on terrain’s topology and material composition. The learning method allows battery-operated robots navigating unfamiliar terrain to achieve high-quality scooping actions using vision and limited on-line training experience. The researchers trained a robot modeled on a lander arm on materials of varying sizes and volumes on 67 different terrains. The NASA Jet Propulsion Laboratory plans to use the model in its Ocean World Lander Autonomy Testbed. UIUC's Pranay Thangeda observed that the lander’s batteries have a lifespan of about 20 days, so “We can't afford to waste a few hours a day to send messages back and forth" to instruct the lander how to proceed.
|
Biological Camera Stores Images
Computer Weekly
Aaron Tan
July 12, 2023
National University of Singapore (NUS) scientists used biological components to encode and store images on live cells. Said NUS' Poh Chueh Loo, "Using optogenetics—a technique that controls the activity of cells with light akin to the shutter mechanism of a camera—we managed to capture 'images' by imprinting light signals onto the DNA 'film'." The researchers marked the images for identification using barcoding methods similar to photo labeling. They utilized machine learning algorithms to structure, sort, and reassemble the images in a manner resembling digital cameras' data capture, storage, and retrieval processes. The system is easily reproducible and scalable compared with previous DNA data storage techniques.
|
This Chatbot Sounds Familiar
USC Viterbi School of Engineering
Julia Cohen
July 10, 2023
The University of Southern California (USC)'s Jonathan May and colleagues designed the RECAP (Retrieval-Enhanced Context-Aware Prefix Encoder for Personalized Dialogue Response Generation) model to enable chatbots to mimic their users based on written text. The researchers used Reddit as a data source for user-written text in developing underlying language models to help choose the likely next word in a conversation. May explained, "You give the model a history of the context of a conversation—everything that's happened up until now—and ask, 'return similar sentences from my past.'" RECAP will produce a response whose content and style are similar to a given author's previous dialogues.
|
Theaters Tempt Audiences with Virtual Reality
BBC News
Michael Dempsey
July 6, 2023
U.K. theaters are experimenting with virtual reality (VR) provided by companies like Box Office VR to entice new audiences. Box Office VR designed its technology to interoperate with inexpensive products like Google Cardboard VR Glasses. Patrons insert their phones into the VR system, then don the headset and plug in their headphones; the headset amplifies a recorded performance streamed to the phone, delivering a 180-degree view of the stage and auditorium. Earlier this year, the Dundee Repertory company offered headsets to patrons to preview a play about Dundee United football team manager Jim McLean, in which they could see things from the actor's perspective in the play's first half.
|
Learning the Language of Molecules to Predict Their Properties
MIT News
Adam Zewe
July 7, 2023
A unified framework developed by researchers at the Massachusetts Institute of Technology (MIT) and the MIT-IBM Watson AI Laboratory can forecast molecular properties while producing new molecules using only small datasets. The researchers programmed a machine learning (ML) system to automatically learn the "molecular grammar" of a domain-specific dataset in order to build viable molecules and anticipate their characteristics "so you can train a model to do the prediction without all of these cost-heavy experiments," according to MIT's Minghao Guo. The team organized a hierarchical strategy to accelerate molecular grammar-learning, decoupling the process into a widely applicable metagrammar provided at the outset and a molecule-specific grammar from the domain dataset. This approach accurately predicted molecular properties and generated viable molecules from a dataset containing fewer than 100 samples, outperforming several popular ML techniques on small and large datasets.
|
Algorithm Breaks Through 'Search Bubbles'
New York University
July 10, 2023
Computer scientists at New York University (NYU) have developed an algorithm that mimics ecosystem interactions to make search-based recommendation systems more diverse and useful than "filter bubbles," which provide very similar or identical search results. Pyrorank, which serves as an "add-on" to existing recommender systems, lessens the impact of content the user has purchased or interacted with when generating recommendations. The researchers used datasets from MovieLens, Good Books, and Goodreads to compare Pyrorank's search outcomes with traditional recommender systems and found Pyrorank broke through filter bubbles to produce more diverse recommendations. While greater diversity in search results could result in less accuracy, NYU's Anasse Bari said, recommender systems “can be calibrated to bolster the heterogeneity of search outcomes while minimizing the loss of accuracy."
|
Chip Makers Stack 'Chiplets' Like Lego Blocks to Drive AI
The Wall Street Journal
Yang Jie
July 10, 2023
Chip makers are stacking pre-existing chips like toy building blocks to speed the development of more powerful chips. IBM’s Darío Gil said, "A huge part of the future of semiconductors is packaging and chiplets. It's just much more powerful than having to design a massive chip from scratch." A coalition of chip makers formed last year aims to develop standards for chiplet design. Chiplet technology has gained popularity among companies looking to quickly design chips for use in artificial intelligence technology, but chiplet production costs remain high, so chiplet designs are viewed as more suitable for higher-end products, and combining several less-advanced chips is seen as a more cost-effective way to boost performance.
*May Require Paid Registration
|
Game-Playing Automaton Acts Like 'Irrational' Human
Cornell Chronicle
Patricia Waldron
July 10, 2023
Cornell University researchers demonstrated that a probabilistic finite automaton will exhibit human behaviors when its memories are limited, supporting the concept of bounded rationality. In a wildlife poaching game in which players are either a rhino poacher or a ranger working to stop poaching, the automaton would execute an optimal game strategy when it was programmed to remember every move, but with its memory limited, the automaton acted like humans in taking decision-making shortcuts, emulating probability matching and overweighting significant results. In observing roughly 100 people playing as poacher, the researchers identified similarities in gameplay between humans and automatons and found the automaton's performance improved when it recreated seemingly "irrational" human behaviors. Cornell's Joseph Halpern said this can be interpreted as "doing the best you can given your computational limitations."
|
AI Helps Translate Bible into Very Rare Languages
The Washington Post
Fiona Andre
July 6, 2023
The University of Southern California's Joel Mathew and Ulf Hermjakob aim to simplify translating the Bible into extremely rare languages through the use of an artificial intelligence (AI)-driven platform called the Greek Room. The platform integrates spell-checking, world alignment for translation consistency, and the Wildebeest application for detecting improper characters in scripts. Mathew said he and Hermjakob designed the Greek Room to focus on "very low-resource languages that are not even in the top 500"; Hermjakob cited Ethiopia's Uyghur dialect and northern Kenya's Oromo dialect as examples of such languages.
*May Require Paid Registration
|
|
