Welcome to the November 13, 2017 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 Pulls Ahead of U.S. in Latest Top500 List
TOP500.org
November 13, 2017
China has overtaken the U.S. in the latest Top500 ranking of supercomputers by a margin of 202 to 143, representing the largest number of Chinese systems on the list and the lowest number of U.S. systems in 25 years. Six months ago, the U.S. led with 169 systems to China's 160. China's Sunway TaihuLight and Tianhe-2 were ranked first and second on the list, with respective High Performance Linpack (HPL) marks of 93.01 petaflops and 33.86 petaflops. All top 10 supercomputers delivered more than 10 petaflops on HPL for the first time, and the aggregate performance of all 500 systems has increased to 845 petaflops, versus 749 petaflops six months ago and 672 petaflops 12 months ago. However, the approximately 100-petaflop boost is well below the Top500's long-term historical trend. The list's entry point rose to 548 teraflops compared to 432 teraflops in June, with the 548-teraflop system ranked 370th on the previous list.
|
IBM Raises the Bar With 50-Qubit Quantum Computer
Technology Review
Will Knight
November 10, 2017
IBM last week announced a 50-quantum bit (qubit) quantum computer, as well as the availability of a 20-qubit system via its cloud computing platform. The researchers say both systems preserve their quantum state for 90 microseconds, which constitutes an industry milestone, although it is still a very short duration. IBM's Dario Gil notes the company's quantum cloud software upgrade shows that "we're at world-record pace. But we've got to make sure non-physicists can use this." Meanwhile, University of Maryland professor Andrew Childs says IBM's 50-qubit breakthrough does not necessarily ensure a major computational jump. "Those qubits might be noisy, and there could be issues with how well connected they are," he notes. Childs also says without studying the details of how the new quantum system functions, it is difficult to validate the researchers' assertions. The IBM team suggests quantum systems bigger than 50 qubits can be modeled on conventional systems using clever mathematical techniques.
*May Require Free Registration
|
Training for Artificial Intelligence in Warfare
U.S. News & World Report
Sintia Radu
November 8, 2017
With technology rapidly evolving, governments are taking a closer look at artificial intelligence (AI) as a way to further their strategic interests in areas such as national defense, which in turn raises ethical questions over issues such as AI's ability to misshape reality and the role of human decision-makers. "We have had this fundamental truth for all history that if you can see it or you can hear it, it is fact," says Booz Allen Hamilton's Steve Mills. "AI can take that away from us." Meanwhile, Katherine Charlet with the Carnegie Endowment for International Peace notes machine learning can enable less sophisticated hackers to orchestrate more refined cyberattacks. Experts think addressing AI's negative national defense implications requires understanding the boundaries and responsibilities humans have when contending with machines. "The challenge is to make sure that [the responsible] person or organization is actually able to control and influence what might go wrong," says Princeton University professor Edward Felten.
|
Walk This Way: A Better Way to Identify Gait Differences
Osaka University
November 7, 2017
Researchers at Osaka University in Japan are developing new input/output architectures for convolutional neural network-based (CNN) cross-view gait recognition by utilizing a Siamese network for verification, in which an input is a pair of gait features for matching, and an output is genuine, indicating the same subjects, or imposter, indicating different subjects. "Current CNN-based approaches are missing the aspects on verification versus identification, and the trade-off between spatial displacement, that is, when the subject moves from one location to another," says Osaka's Noriko Takemura. The researchers say the Siamese network architectures resist spatial displacement, as the difference between a matching pair is calculated at the last layer after passing through the convolution and max pooling layers. "We conducted experiments for cross-view gait recognition and confirmed that the proposed architectures outperformed the state-of-the-art benchmarks in accordance with their suitable situations of verification/identification tasks and view differences," says Osaka's Yasushi Makihara.
|
Researchers Developing Building-Free Data Centers
Network World
Patrick Nelson
November 8, 2017
Researchers at Horizon Computing in France are co-developing an outdoor, building-free server farm using vats of liquid to cool computers. The containers of non-conductive mineral oil, known as RuggedPODs, could conceivably be installed in the middle of fields, and other benefits to the containers cited by Horizon include the ability to function in extreme environments and positive temperatures. Horizon also thinks RuggedPOD costs could be kept low because the containers are easy to manufacture. The company says each container should be able to store 32 to 48 central-processing unit cores and run on an 800-Watt power supply. Horizon envisions hundreds of RuggedPODs in automated outdoor data centers, and the current prototype combines a rail and hoist-driven system to provide maintenance. "This automated data center has to be fully functional no matter if it rains, snows, if it is windy, hot, or cold, and with as little as possible human maintenance," the researchers say.
|
Research Drive Could 'Untangle' Vexing Problem in Computer-Simulation Technology
KU News
Brendan M. Lynch
November 8, 2017
Researchers at the University of Kansas (KU) and the University of Wisconsin-Madison have found the use of a tangled mesh in a finite element simulation can lead to inaccurate tests, with potentially disastrous consequences in biomechanical design, product development, or large-deformation analysis. The researchers are using a U.S. National Science Foundation grant to explore new methods for addressing the tangled mesh problem. The team will create new constrained optimization methods for mesh untangling to convert "severely tangled meshes into mildly tangled meshes," and it will refine the finite-cell method to ensure accurate finite-element solutions over these mildly tangled meshes. The KU researchers will develop a method to untangle meshes so they can be used with standard finite elements, while the University of Wisconsin-Madison researchers will develop a finite-element solver that can work on tangled meshes. In addition, the researchers will exchange teaching modules in the form of prerecorded lectures to be used in graduate-level classes.
|
Navy Using New UConn Software to Improve Navigation
UConn Today
Kristen Cole
November 7, 2017
Researchers at the University of Connecticut (UConn) have developed a prototype of the Tool for Multi-objective Planning and Asset Routing (TMPLAR), which is now being used by the U.S. Navy to enhance the ability of ships to reroute through unpredictable weather. TMPLAR is integrated with the Navy's meteorology and oceanographic weather forecasts, and its algorithms account for variables such as ocean depth, undersea pipelines, cables, and oil rigs. They also factor in multiple user goals, whether to traverse to an area to minimize travel time, maximize fuel efficiency given the predicted weather, meet training objectives, or maximize operational endurance. "The tool guarantees safe travel from any point in the ocean, above, on, or below its surface, while making choices en route that optimize fuel consumption and cater to any set of objectives of the operator," says UConn's David Sidoti. "Using special clustering techniques, the tool's algorithms have even been applied to finding low-risk routes that avoid storms or hurricanes."
|
We've Figured Out How to Ensure Quantum Computers Can Be Trusted
New Scientist
Mark Kim
November 7, 2017
Researchers at the University of Maryland and the Georgia University of Technology have demonstrated a quantum program that detects data corruption. In traditional computers, error detection and correction are done with duplicated data, and mistakes are fixed by reconstructing the erroneous bits from uncorrupted aspects of the system. In quantum computers, it is impossible to duplicate quantum states without measuring them, and measurement causes information loss. The new solution involves five quantum bits (qubits), meaning for every two qubits' worth of information, there are four possible combinations. The new program uses four qubits to record these states, while the fifth qubit catches errors in the other four. If there is an error in one of the qubits, the fifth qubit will note the uneven distribution of 1s or 0s and change its state. The new verification system reduces the error rate from up to 15 percent down to 0.1 percent, according to the University of Maryland's Norbert Linke.
|
Researcher Seeks to Tame 'Ghost' of Uncertainty in Complex Dynamic Systems
KU News
Brendan M. Lynch
November 7, 2017
University of Kansas professor Huazhen Fang is leading research to address the "unstructured uncertainty" challenge in the quest to develop accurate predictive mathematical equations and algorithms dealing with complex dynamic systems. "We try to identify the probability-based appearance of uncertainties conditioned on the data," Fang says. "This will lead us to develop mathematical models and efficient algorithms that can effectively account for the ghost presence of uncertainty." Fang's efforts concern holistic investigation, which not only deals with predicting a system's behavior despite uncertainties but also analyzes sensor-based observation structure to acquire high-quality data beneficial for prediction. Fang says the project could have many military uses, including enabling more robust autonomy, navigation, guidance, target tracking, sensor fusion, fault detection, and structural health prognostics. He also notes the basic research could power consumer products such as global-positioning system navigation by augmenting the Kalman filter, an algorithm already used to contend with uncertainty.
|
Why They Dox: First Large-Scale Study Reveals Top Motivations and Targets for This Form of Cyber Bullying
NYU Tandon School of Engineering
November 7, 2017
Researchers at New York University (NYU) and the University of Illinois at Chicago (UIC) say they have published the first large-scale study of doxing, a low-tech, high-harm form of online harassment that involves collecting and publishing sensitive personal information online to exact revenge, seek justice, or intimidate victims. The researchers created a custom text classifier that enabled them to identify and analyze dox files, and they found doxing victims are more likely than others to close or increase the privacy settings of social media accounts following an attack. The team also found that revenge and justice are the primary motivators for doxing attacks, while competition and politics comprise only 1 percent each of the reasons identified by the study. "The ability to detect doxing and identify the primary motivations for these attacks is key to helping Internet service providers, law enforcement, and social media networks better protect users from harassment," says NYU professor Damon McCoy.
|
Machine Learning and Deep Learning Programs Provide a Helping Hand to Scientists Analyzing Images
Phys.org
November 9, 2017
Researchers at the U.S. Department of Energy's Oak Ridge National Laboratory (ORNL) and Fermilab are using machine learning and deep learning to better identify how neutrinos interact with normal matter. "[Machine-learning] techniques are extremely efficient at finding subtle signals" such a small shifts in particle tracks, notes Fermilab's Gabe Perdue. To prepare for when the Large Synoptic Survey Telescope goes online in 2022, researchers at SLAC National Accelerator Laboratory for the first time used deep learning to analyze complex distortions in spacetime, which is 10 million times faster than traditional analytic methods, while boasting the same accuracy. Meanwhile, Lawrence Berkeley National Laboratory researchers have developed the deep-learning pyCBIR tool to help them match images to similar ones already in the lab database. The ORNL/Fermilab team also is designing a new deep-learning program that mates quantum-data computer processing, supercomputing, and brain-like hardware together to produce highly accurate data analysis.
|
Deep Learning for Science: A Q&A With NERSC's Prabhat
HPCwire
Kathy Kincade
November 7, 2017
In an interview, the U.S. National Energy Research Scientific Computing Center's (NERSC) Prabhat discusses deep learning, machine learning, and the challenges of applying them to science. "I think of deep learning...as a subset of machine learning, which in turn is closely related to the field of statistics, and all of them have to do with solving inference problems of one kind or another," Prabhat says. He cites the accessibility of big data, more powerful computers, and their convergence as driving deep learning forward, and notes the preponderance of convolutional network and long short-term memory architectures in deep-learning scientific applications. Prabhat says NERSC has about 70 users employing deep-learning software, and it has identified and deployed several popular deep-learning frameworks on its Cori system. The challenges Prabhat foresees for deep learning include handling complex and diverse datasets, performance and scaling, and addressing "a lack of theory, interpretability, uncertainty quantification, and the need for a formal protocol."
|
|
