Welcome to the November 23, 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."

China’s flag. China's Surveillance State Sucks Up Data. U.S. Tech Is Key to Sorting It
The New York Times
Paul Mozur; Don Clark
November 22, 2020


Chips from U.S. companies Intel and Nvidia power a Chinese supercomputing facility that monitors people in a region known for government suppression, raising issues about the technology sector's responsibility. The Urumqi Cloud Computing Center watches the population of Xinjiang using Intel and Nvidia chips sold to Sugon, the company backing the center, to sort the collected data. Local officials in 2017 said the center would support a Chinese police surveillance project capable of searching 100 million photos a second, and by 2018 its systems could link to 10,000 video feeds and analyze 1,000 simultaneously, via artificial intelligence. Intel and Nvidia claimed they were unaware of what they termed misuse of their technology.

Full Article
*May Require Paid Registration
Time Test Measures How Fast a Neural Net Can Be Trained
ZDNet
Tiernan Ray
November 18, 2020


Computer industry consortium MLPerf has launched a series of test results for high-performance computing systems running machine learning tasks, designed to quantify a deep learning network's training time on the world's most powerful computers. The tests measure how many minutes the supercomputers take to train the network until it becomes proficient in the CosmoFlow and DeepCAM tasks. The supercomputer tasks count wall time to achieve accuracy, where less, rather than more, is better. Japan's AI Bridging Cloud Infrastructure Computer (ABCI) determined 13 minutes to be the fastest CosmoFlow training time, while a second ABCI system completed DeepCAM training fastest, in 10.5 minutes.

Full Article

RIT researchers hope to speed up IP lookup times. RIT Researchers Create Programmable Network Switch to Improve IP Lookup Time
Rochester Institute of Technology
Scott Bureau
November 19, 2020


Rochester Institute of Technology (RIT) researchers have developed a prototype programmable switch that shortens Internet Protocol (IP) lookup time and upgrades router performance dramatically. The RIT team used a novel longest prefix matching algorithm (CuVPP) which employs packet batch processing and cache locality for both instructions and data by exploiting Vector Packet Processing (VPP). The switch also taps a cuckoo filter in the cache as a fast pre-screening mechanism. The researchers said CuVPP can realize up to 4.5 million lookups per second with real traffic, outperforming other popular strategies. Cisco Systems' John Marshall said, "As the Internet continues to grow, technology to improve the efficiency of IPv6 is necessary. This algorithmic technique can really improve packet forwarding performance in software routers."

Full Article

A modern kitchen. The Future of Kitchen Design Is Hands-Free, Smartphone-Activated
The Wall Street Journal
J.S. Marcus
November 18, 2020


Kitchen appliance designers and manufacturers increasingly rely on technological advances, using cameras, sensors, artificial intelligence, and novel materials to modernize the kitchen. Appliance maker Miele's G7000 dishwashers are equipped with sensors that prompt owners' smartphones to reorder detergent when running low, while rival Bosch's dishwashers are made with a Zeolite mineral compound that helps plastic items dry more efficiently. Bosch also has integrated refrigerator cameras with its new cloud-accessing system, so they can suggest recipes based on ingredients currently stocked. The Covid-19 pandemic also is reshaping homeowners' kitchen expectations, with touchless features and sanitization increasingly desired.

Full Article
*May Require Paid Registration
3D-Printed Glass Enhances Optical Design Flexibility
Lawrence Livermore National Laboratory
Anne M. Stark
November 18, 2020


Researchers at the U.S. Department of Energy's Lawrence Livermore National Laboratory (LLNL) used multi-material three-dimensional (3D) printing to produce tailored gradient refractive index (GRIN) glass optics. The LLNL team controlled the ratio of two distinct glass-forming pastes (inks) combined inline via the direct ink writing method of 3D printing. Tailoring the GRIN enables the replacement of a curved optic with a flat surface, which could lower finishing costs and add surface curvature to manipulate light using bulk and surface effects. LLNL's Rebecca Dylla-Spears said, "This is the first time we have combined two different glass materials by 3D printing and demonstrated their function as an optic. Although demonstrated for GRIN, the approach could be used to tailor other material or optical properties as well."

Full Article

Two companies have partnered to develop computer vision for e-scooters. E-Scooters Getting Computer Vision to Curb Pedestrian Collisions
The Washington Post
Dalvin Brown
November 19, 2020


Micromobility companies Luna and Voi Technology partnered to launch a test fleet of electric scooters (e-scooters) that use computer vision to avoid collisions with pedestrians in the U.K. city of Northampton. Luna developed a system of cameras and sensors that reportedly will enable the scooters to learn and respond to their environments. Luna's Ronan Furlong said the pedestrian detection system was trained on thousands of pictures of real people, while the data from the vehicles can be used to slow an e-scooter down, minimize congestion, or bar riders who are non-compliant with local transportation rules. Swedish e-scooter manufacturer Voi incorporated Luna's computer-vision system into 50 of its vehicles. The deployment will include access to a real-time camera feed so local authorities can monitor and police rider behavior.

Full Article
*May Require Paid Registration
World's Smallest Atom-Memory Unit Created
UT News
November 19, 2020


Engineers at the University of Texas at Austin created the smallest atom-memory device (atomristor) to date by working out the physics mechanism that unlocks dense memory storage capabilities for these devices. The atomristor, designed using facilities at the U.S. Department of Energy's Oak Ridge National Laboratory, promises capacity of about 25 terabits per square centimeter, which surpasses commercially available flash memory devices' per-layer memory density 100-fold. Said the university's Deji Akinwande, "The scientific holy grail for scaling is going down to a level where a single atom controls the memory function, and this is what we accomplished in the new study."

Full Article
Quantum Tunneling Pushes Limits of Self-Powered Sensors
The Source (Washington University in St. Louis)
Brandie Jefferson
November 16, 2020


Washington University in St. Louis (WUSTL) researchers have developed self-powered quantum sensors that utilize quantum tunneling, and can operate on their own for more than a year following a small initial energy input. The WUSTL team combined four capacitors and two transistors into two dynamical systems—a reference and sensing system—featuring two capacitors and a transistor each. Each capacitor holds a small initial charge of about 50 million electrons, and adding a transducer to one system enabled the measurement of ambient micromotion with a piezoelectric accelerometer; shaking the accelerometer converts that motion into electricity, which reconfigures the Fowler-Nordheim tunneling barrier. The WUSTL researchers read voltage in both the sensing and reference system capacitors, and used the difference in voltages to calculate true electron-tunneling rate measurements from the transducer. WUSTL's Shantanu Chakrabartty said, "As long as you have a transducer that can generate an electrical signal, it can self-power our sensor-data-logger."

Full Article
Upgraded Radar Can Enable Self-Driving Cars to See Clearly, No Matter the Weather
UC San Diego News Center
Liezel Labios
November 17, 2020


Electrical engineers at the University of California, San Diego (UC San Diego) have come up with a way to enhance the imaging capability of radar sensors to accurately predict the shape and size of objects in the vicinity. This could enable self-driving cars to navigate safely in any type of weather, using what UC San Diego's Dinesh Bharadia calls "a LiDAR-like radar." The system features two radar sensors positioned on the hood 1.5 meters (4.9 feet) apart, to capture more space and data than a single sensor can; the UC San Diego team developed algorithms that combine the collected data into a single noise-free image. Test drives in clear weather showed the system performed as well as a LiDAR sensor in ascertaining the dimensions of cars moving in traffic, with similar performance in simulated foggy weather.

Full Article

A Parrot Swing drone avoiding obstacles. ML Guarantees Robots' Performance in Unknown Territory
Princeton Engineering News
Molly Sharlach
November 17, 2020


Princeton University researchers have developed a machine learning (ML) technique for ensuring robots' safety and success in unfamiliar environments. The researchers came up with the technique by adapting ML frameworks from other fields to robotic movement and grasping. The new technique was tested in various simulations, and also validated by evaluating its use for obstacle avoidance using a small combination quadcopter/fixed-wing airplane drone that flew down a 60-foot-long corridor dotted with cardboard cylinders; it avoided those obstacles 90% of the time. The Toyota Research Institute's Hongkai Dai said, " Over the last decade or so, there’s been a tremendous amount of excitement and progress around machine learning in the context of robotics, primarily because it allows you to handle rich sensory inputs,” like images captured by a robot’s camera.

Full Article
ACM Transactions on Data Science
 
ACM Publications
 

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]