Welcome to the June 24, 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."
|
|
|
AI Researchers Condemn Predictive Crime Software, Citing Racial Bias, Flawed Methods
TechCrunch
Taylor Hatmaker
June 24, 2020
A coalition of more than 1,000 researchers, academics, and experts in artificial intelligence condemned soon-to-be-published research claims of predictive crime software. The opponents sent an open letter to the publisher Springer, asking that it reconsider publishing the controversial research. Authors and Harrisburg University researchers Roozbeh Sadeghian and Jonathan W. Korn claim their facial recognition software can forecast whether a person will become a criminal, but the coalition expressed doubts on their findings, citing "unsound scientific premises, research, and methods, which numerous studies spanning our respective disciplines have debunked over the years." The letter from the coalition said, "The uncritical acceptance of default assumptions inevitably leads to discriminatory design in algorithmic systems, reproducing ideas which normalize social hierarchies and legitimize violence against marginalized groups."
|
Japanese Supercomputer Is Crowned World's Speediest
The New York Times
Don Clark
June 22, 2020
The Fugaku supercomputer in Japan was ranked the world's fastest in the biannual Top500 list, dethroning an IBM system at Oak Ridge National Laboratory (ORNL). Installed in the city of Kobe by the RIKEN Institute, Fugaku executes 2.8 times more calculations per second than the ORNL system. RIKEN said Fugaku is being used to help study, diagnose, and treat Covid-19; its predecessor, RIKEN's K Supercomputer, gained the lead spot on the Top500 list in 2011. Fugaku was co-developed by technology company Fujitsu, utilizing chips designed with technology from semiconductor and software maker Arm. Arm's Christopher Bergey said Fugaku "is the culmination of almost 10 years of investment and work."
*May Require Paid Registration
|
Researchers Develop Tool to Protect Children's Online Privacy
UT Dallas News Center
Kim Horner
June 23, 2020
Researchers at the University of Texas at Dallas (UT Dallas), the Georgia Institute of Technology, New York University, and Intel have developed a tool that can determine whether mobile applications for children comply with the federal Children's Online Privacy Protection Act (COPPA). The researchers used their COPPA Tracking by Checking Hardware-Level Activity (COPPTCHA) tool to determine that 72 of 100 mobile apps for children that they examined violated COPPA. COPPTCHA accesses a device's special-purpose register, a temporary data-storage site within a microprocessor that tracks its activities, and detects the signature of an app transmitting data. The tool was found to be 99% accurate in assessing apps' COPPA compliance; it found many popular game apps for young children exposed users' Android IDs, Android advertising IDs, and device descriptions. UT Dallas' Kanad Basu said apps that violate COPPA pose privacy risks that could enable bad actors to ascertain a child's identity and location.
|
Model Helps Robots Think More Like Humans When Searching for Objects
The Michigan Engineer News Center
June 18, 2020
University of Michigan (U-M) researchers have developed a model for a practical technique that robots can use to visually search for or target items in complex environments in a more humanlike manner. The Semantic Linking Maps (SLiM) model teaches robots to seek items in close proximity if they are already in sight of a "landmark object." SLiM links certain landmark objects in the robot's memory to other related objects, along with data about the two objects' typical spatial relationships. The researchers employed SLiM to factor in features of both target and landmark objects, in order to give robots a stronger understanding of how things can be arranged in an environment. Said U-M’s Zhan Zeng, “Being able to efficiently search for objects in an environment is crucial for service robots to autonomously perform tasks. We provide a practical method that enables robot to actively search for target objects in a complex environment.”
|
Researchers Use Simulation to Teach Drones to Catch Objects
Venture Beat
Khari Johnson
June 18, 2020
Researchers at the Allen Institute of Artificial Intelligence and the University of Washington trained a drone agent with a box on top to catch a variety of objects in a simulated environment. In a simulation set in a living room in the AI2-THOR photo-realistic simulated environment, objects were thrown 6.5 feet toward a drone agent; the drone's success rate in catching thrown objects ranged from 0% with toilet paper, to 64.4% with toasters (other objects virtually thrown included alarm clocks, heads of lettuce, books, and basketballs). The model was trained using a data set of 20,000 throws, with the launcher randomly positioned for each throw.
|
CMU Method Makes More Data Available for Training Self-Driving Cars
Carnegie Mellon University School of Computer Science
Byron Spice
June 17, 2020
Carnegie Mellon University (CMU) researchers developed a technique for training self-driving cars' tracking systems more effectively by providing larger road and traffic datasets. Previous methods for training autonomous cars' LiDAR systems have used labeled datasets, or sensor data annotated to track each three-dimensional point over time. The expense and labor of manual labeling often necessitates scene-flow training with simulated data, refined by a small volume of labeled real-world data. The simpler CMU approach uses unlabeled data generated by mounting LiDAR on a vehicle and driving around, with the system detecting its own errors in scene flow. Scene-flow accuracy using a training set of synthetic data was only 25%, which rose to 31% when adding a small amount of real-world labeled data, and then to 46% when adding a large amount of unlabeled data.
|
Uber, Lyft Pricing Algorithms Charge More in Non-White Areas
New Scientist
Donna Lu
June 18, 2020
George Washington University (GW) researchers discovered apparent racial bias in pricing algorithms used by ride-hailing companies like Uber and Lyft. GW's Aylin Caliskan and Akshat Pandey analyzed transport and census data in Chicago and found that the firms charge higher fares if a pick-up point or destination is in a neighborhood with a greater ethnic minority population than for those with predominantly white residents. The researchers compared trip data from ride-hailing apps to information from the U.S. Census Bureau's American Community Survey. Although rider ethnicity is excluded from the trip data, fare prices per mile were higher overall if the pick-up or drop-off location was in a neighborhood with fewer white residents, a lower median house price, or lower average educational level. Os Keyes at the University of Washington in Seattle said, "This should cause us to further question studies of 'fairness' and 'bias' in algorithms which promise to end algorithmic racism by simply not mentioning race."
|
This Dangerous Keylogger Could Change the Entire Malware Space
TechRadar
Anthony Spadafora
June 15, 2020
Cofense Intelligence, a developer of “intelligent phishing defense solutions,” is raising concerns about a new keylogger because of how quickly the malware is updated. According to Cofense, the creator of the Mass Logger keylogger, NYANxCAT, has been quickly adding features in response to customer feedback, with 13 updates seen over a recent three-week period. Cofense detected a campaign that delivered an encrypted Mass Logger binary using an attached GuLoader executable. Cofense also found NYANxCAT has incorporated advanced features into Mass Logger, such as its USB spreading capability and a function that allows cybercriminals to search for files with a specific file extension and exfiltrate them. NYANxCAT indicated in patch notes that new targets were added for the keylogger's credential stealing functionality, and measures were taken to reduce automated detection. Cofense said network admins should keep an eye out for FTP sessions or emails sent from local networks that do not conform to their organization's standards.
|
Deep Learning-Based Surrogate Models Could Hasten Scientific Discoveries
Lawrence Livermore National Laboratory
June 17, 2020
Lawrence Livermore National Laboratory (LLNL) researchers have designed neural network-based surrogate models that can outperform computationally expensive simulators. The team developed a deep learning-driven Manifold & Cyclically Consistent (MaCC) surrogate model incorporating a multi-modal neural network for accurately mimicking complex scientific processes, including high-energy density physics involved in inertial confinement fusion (ICF). Applying MaCC to ICF implosions conducted at LLNL’s National Ignition Facility showed it could adequately replicate the simulator and outperform state-of-the-art surrogate models across a broad spectrum of metrics. LLNL's Timo Bremer said the surrogate model's ability to analyze a large volume of complex data has ramifications for stockpile modernization, and could potentially lead to new scientific discoveries and a novel class of simulation performance and analysis methods.
|
Design Allows Computer Engineers to Mix Systems to Boost Performance
Princeton Engineering News
Adam Hadhazy
June 18, 2020
Princeton University researchers have built a hardware platform that allows different types of computer cores to fit together to boost performance, enabling new kinds of system customization by computer engineers. The objective is to design systems that distribute tasks among specialized cores, boosting efficiency and speed, without relying on a single Instruction Set Architecture (ISA). The Bring Your Own Core (BYOC) platform operates on two basic levels: the first is a detailed hardware model, and the second is an emulation of the chip framework run on reprogrammable hardware, which approximates the appearance and function of a real computer chip using select cores and ISAs. The open source BYOC allows researchers to link their cores into a modifiable hardware architecture, which supports multiple ISAs and can scale up to 500 million cores.
|
Technique May Enable All-Optical Datacenter Networks
University College London
June 22, 2020
A study by researchers at University College London (UCL) and Microsoft Research Cambridge in the U.K. demonstrates a new technique to potentially advance the deployment of all-optical datacenter networks. The process involves synchronizing the clocks of all connected servers via optical fiber, and programming hardware to memorize clock-phase values so clock time does not require re-checking, practically eliminating the time required to "recover" the clock. This clock-phase caching method could synchronize the clocks of thousands of computers in less than a billionth of a second. The researchers showed that shortening clock recovery time to under a nanosecond significantly boosted the performance of optical switching, compared to state-of-the-art solutions. UCL's Kari Clark said, "It has the potential to transform communication between computers in the cloud, making key future technologies like the Internet of Things and artificial intelligence cheaper, faster, and consume less power."
|
'Low-Cost Android' to Study the Brain
CORDIS
June 17, 2020
Researchers at Europe's MoCoTi (Motor Control and Timing in the Cerebellum: Spatio-Temporal Integration in Complex Neuronal Networks) and Myorobotics initiatives have designed a prototype "low-cost android" that learns how to actuate its own limbs and is easily reproducible. The device features an artificial cerebellum to control a tendon-driven robotic arm, with a modular design to enable relatively efficient mass production. The researchers used the Myorobotics system, in which nine muscles comprised of mechatronic devices are coordinated to drive spherical articulation, with one biceps-connected device joined to two articulations linking the shoulder to the elbow. The artificial cerebellum duplicates key neurons, their connectivity, and their adaptation and learning in a real-time simulation, via the SpiNNaker neuromorphic computer at the U.K.'s University of Manchester.
|
AI Takes Player Performance Analysis to New Dimension
Loughborough University
June 18, 2020
Computer scientists at Loughborough University in the U.K. have developed artificial intelligence algorithms that could revolutionize player performance analysis for football (soccer) clubs. The researchers designed a hybrid system that accelerates and supplements human data entry with camera-based automation to meet demand for timely performance data generated from large amounts of videos. The team applied the latest computer vision and deep learning technologies to identify actions by detecting players' body poses and limbs, and trained the deep neural network to track individual players and capture data on individual performance throughout the match video. Loughborough’s Baihua Li said the new technology “will allow a much greater objective interpretation of the game as it highlights the skills of players and team cooperation.”
|
|
