Welcome to the November 4, 2020 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."

A voter sports patriotic shoes in California. Election officials across the U.S. are also putting their sneakers to work as they hand-carry voting results to avoid cyber risks. 'Sneakernet' Helps Election Officials Process Results
The Wall Street Journal
Jared Council; Sara Castellanos; John McCormick
November 3, 2020


U.S. election officials used voting machines and other devices linked to the "sneakernet"—a system for transmitting electronic data by physically conveying it from place to place—to process results from Tuesday's elections. Much of the tabulating, reporting, and auditing process is digitized, dependent on specialized software and computers not connected to the Internet, to thwart hacking. Voting data from scanned paper ballots is extracted onto flash drives hand-carried to central locations to tabulate and report results. The drives also are used to send total vote tallies to a central, Internet-connected computer. Election experts said the sneakernet is a much safer means for sharing data than online. Said Barbara Simons, chair of the board of directors at nonprofit Verified Voting and former ACM president, “We can’t trust computers alone. We need hand-marked paper ballots, systems for voters with disabilities, a strong chain of custody, and postelection ballot audits.”

Full Article
*May Require Paid Registration
Porcupine Molecular Tagging Scheme Offers Sharp Contrast to Conventional Inventory Control Systems
University of Washington Paul G. Allen School of Computer Science & Engineering Allen School News
November 3, 2020


An end-to-end molecular tagging system created by researchers at the University of Washington (UW) and Microsoft can be programmed and read within seconds using a portable nanopore device. The Porcupine system can substitute dehydrated strands of synthetic DNA called molbits (molecular bits) for bulky plastic or printed barcodes, eliminating the need for specialized laboratories and equipment. Porcupine enables the binary 0s and 1s of a digital tag to signal the presence or absence of each of 96 molbits; users read a tag's data by rehydrating and running it through Oxford Nanopore Technologies' MinION device. UW's Luis Seze said, “Porcupine is one more exciting example of a hybrid molecular-electronic system, combining molecular engineering, new sensing technology, and machine learning to enable new applications.”

Full Article
Computer Vision Algorithms Pretrained on ImageNet Exhibit Multiple, Distressing Biases
VentureBeat
Kyle Wiggers
November 3, 2020


Scientists at Carnegie Mellon and George Washington Universities said they have unearthed worrying biases in state-of-the-art computer-vision algorithms pretrained on the ImageNet online photo dataset. The researchers used their proposed Image Embedding Association Test (iEAT) benchmark to adapt word embedding tests to compare pooled image-level embeddings, in order to quantify biases embedded during unsupervised pretraining. The team said experiments with the ImageNet-pretrained iGPT algorithm from OpenAI and Google's SimCLRv2 indicated that both contain "significant" biases, including racial prejudices and stereotyping, and gender and weight biases, while iGPT's next-level prediction discriminated against women. The researchers said, "Patterns of stereotypical portrayal of social groups do affect unsupervised models, so careful research and analysis is needed before these models make consequential decisions about individuals and society."

Full Article

Roberta Pirazzini prepares a drone for flight over the Arctic. First Drone Goes Flying to North Pole on Climate Mission
Bloomberg
Laura Millan Lombrana
October 28, 2020


Researchers at the Finnish Meteorological Institute joined the biggest Arctic expedition in history to fly the first drone near the North Pole to measure surface albedo, or sunlight reflected by the ice. This could help scientists understand how much solar radiation is absorbed by the Earth and how much is reflected back into the atmosphere, in hopes of predicting how fast sea ice will melt. The study was especially challenging due to the deep cold and the fact that global positioning satellites experience small uncertainties at extreme northern latitudes. Despite these and other challenges, the researchers conducted 18 flights of up to 30 meters over the ice over a three-week period.

Full Article
Computational Methods to Ease Reuse of Construction Components
EPFL (Switzerland)
October 30, 2020


Researchers in the Structural Xploration Lab of the Swiss Federal Institute of Technology Lausanne (EPFL) have developed algorithms to help architects incorporate both new and reused components into building designs to reduce their environmental impact. The software application containing the algorithms brings life-cycle assessments into the design process, with a focus on reusing existing steel beams, columns, and bars, as well as wood and concrete. Users input the structure's overall characteristics and its stock of reusable components into the software, which performs an initial optimization of the structure's form to use as little material as possible; it then provides alternative forms that meet sustainability objectives by, for instance, changing the structure's layout or positioning elements from existing materials. The software also identifies the optimal combination of new and recycled components.

Full Article
AI Tool Provides More Accurate Flu Forecasts
Stevens Institute of Technology
November 2, 2020


An influenza forecasting tool powered by artificial intelligence (AI), developed by researchers at the Stevens Institute of Technology, incorporates location data to deliver up to 11% more accurate predictions than other state-of-the-art forecasting techniques, predicting flu outbreaks up to 15 weeks in advance. Stevens' Yue Ning and colleagues used a graph neural network to encode flu infections as interconnected regional clusters, enabling the algorithm to parse out patterns in how infections spread between regions, and to use patterns spotted in one region to inform forecasts in other areas. The team trained the AI tool on real-world state and regional data from the U.S. and Japan, then tested its forecasts against historical flu data. Said Ning, "Our algorithm will keep learning and improving as we collect new data, allowing us to deliver even more accurate long-term predictions."

Full Article
Computational Tools Open New Era of Fossil Pollen Research
National Science Foundation
November 2, 2020


A team including researchers from the Smithsonian Tropical Research Institute, the universities of Illinois at Urbana-Champaign and California, Irvine, and other institutions combined machine learning (ML) with high-resolution imaging to enhance fossil pollen research. The scientists designed and trained three ML models to distinguish between existing Amherstieae legume genera, and tested them against fossil specimens from western Africa and northern South America dating back to the Paleocene era. The models classified existing pollen accurately more than 80% of the time, and strongly agreed on the identification of fossil pollen specimens. Said the National Science Foundation’s Jie Yang, "Machine learning and computer vision technologies can not only lead to new scientific discoveries, but also help us better understand what happened in the past."

Full Article
Rapid Prototyping: Testing Heavy Equipment in Software
Fraunhofer-Gesellschaft
November 2, 2020


Researchers at the Fraunhofer Institute for Industrial Mathematics in Germany have developed a hardware-in-the-loop (HiL) platform that aims to make machine development faster and more affordable through the use of software simulation. Said Fraunhofer's Christian Salzig, "Our HiL simulator allows us to test heavy equipment of all kinds, including a variety of crane types and concrete pumps. This allows us to help optimize prototypes." The process involves reproducing the machine to be tested as a software model, incorporating its technical specifications and the physical laws of mechanics, hydraulics, and electronics as mathematical equations. The simulator with the digital twin is then connected to the electronic control units that control machine operation, with the machine's movements reproduced on an animated, three-dimensional display. The software can simulate the precision of operations between the machine and the control unit, as well as malfunctions and critical borderline situations.

Full Article

This robot uses a subsumption architecture to exhibit more realistic eye gaze. Disney Research Makes Robotic Gaze Interaction Eerily Lifelike
IEEE Spectrum
Evan Ackerman
November 2, 2020


A team of researchers from Disney Research, the California Institute of Technology, the University of Illinois at Urbana-Champaign, and Walt Disney Imagineering is imbuing animatronic robots with lifelike eye gaze. The system they are using decides where to gaze by first identifying a person to target using an RGB-D camera; if multiple people are visible, the system calculates a curiosity score for each, based on the amount of motion, and chooses the highest-scoring target. The robot will then display high-level gaze behavior (reading, glancing, engaging, or acknowledging) determined by score. An underlying subsumption architecture dictates lower-level motion behaviors like breathing, small head movements, eye blinking, and saccades.

Full Article

The underwater navigation system being put into the water An Underwater Navigation System Powered by Sound
MIT News
Daniel Ackerman
November 2, 2020


Massachusetts Institute of Technology (MIT) researchers built a battery-free underwater navigation system that reflects modulated signals from its environment in order to provide positioning information without expending energy. Underwater Backscatter Localization (UBL) employs piezoelectric sensors that produce their own electricity in response to mechanical stress, and use that power to selectively reflect soundwaves. A receiver translates the backscatter sequence into a pattern of 1s and 0s that conveys information about ocean temperature or salinity. To overcome the problem of myriad surface and seabed reflections complicating localization, the observation unit transmits a signal sequence across a range of frequencies, each at a different wavelength; the reflected sound waves return to the unit in different phases, with combined timing and phase data pinpointing distances to the tracking device. MIT’s Fadel Adib said, “We’re hoping to understand the ocean at scale. It’s a long-term vision, but that’s what we’re working toward and what we’re excited about.”

Full Article
Software Can Spy on What You Type in Video Calls by Tracking Your Arms
New Scientist
Chris Stokel-Walker
October 30, 2020


Researchers at the University of Texas at San Antonio developed a model that can track the movement of the shoulders and arms of a person typing during a video call, to determine what they are typing. The person's movements are mapped onto a keyboard using optical flow, and the results are cross-referenced against a dictionary of commonly typed words. The model correctly identified the word being typed 75% of the time, though its success rate varies based on a user's typing skills. For instance, the model correctly identified 83% of words typed by those who "peck" at the keyboard. Further, 3.4% more words were recovered on Skype calls than on Zoom calls, possibly due to the way each app compresses video. The university's Murtuza Jadliwala said users can protect their privacy by blurring their backgrounds, skipping frames in the video, and pixelating their shoulders and arms.

Full Article
*May Require Paid Registration

A White Castle team member standing next to the Flippy robot. White Castle Plans to Use Robot in More Locations
Forbes
Lana Bandoim
October 28, 2020


The White Castle restaurant chain said it is expanding the use of the Flippy Robot-on-a-Rail (ROAR) system to more U.S. locations, after a successful test pilot in one establishment. Developer Miso Robotics said the robot can glide across multiple workstations on a rail, easing storage and use, while relying on artificial intelligence and machine learning to cook food. Miso said, "Capable of identifying current temperatures, predicting time remaining to meet consistency in taste perfection, and alerting staff of unsafe internal cooking temperatures, ROAR features powerful image processing and depth perception from Intel RealSense technology." White Castle intends to use Flippy as a kitchen assistant during night shifts.

Full Article
Tricking Fake News Detectors With Malicious User Comments
Penn State News
Jordan Ford
October 30, 2020


Researchers at the Pennsylvania State University (Penn State) have demonstrated how fake news detectors, like those used by Twitter and Facebook, can be manipulated through user comments. The researchers found that adversaries are able to use random accounts on social media to post malicious comments to flag real stories as fake news or promote fake stories as real news. This involves attacking the detector itself, rather than the story's content or source. The framework developed by the researchers to generate, optimize, and add malicious comments to articles successfully tricked five of the leading neural network-based fake news detectors more than 93% of the time. Penn State's Thai Le said the research "highlights the importance of having robust fake news detection models that can defend against adversarial attacks."

Full Article
ACM Open Access
 
ACM Queue Case Studies
 

Association for Computing Machinery

1601 Broadway, 10th Floor
New York, NY 10019-7434
1-800-342-6626
(U.S./Canada)



ACM Media Sales

If you are interested in advertising in ACM TechNews or other ACM publications, please contact ACM Media Sales or (212) 626-0686, or visit ACM Media for more information.

To submit feedback about ACM TechNews, contact: [email protected]