Association for Computing Machinery
Welcome to the August 7, 2015 edition of ACM TechNews, providing timely information for IT professionals three times a week.

Updated versions of the ACM TechNews mobile apps are available for Android phones and tablets (click here) and for iPhones (click here) and iPads (click here).

HEADLINES AT A GLANCE


Japan Takes Top Three Spots on Green500 List
HPC Wire (08/04/15) Tiffany Trader

Japanese supercomputers have captured the top three positions on the latest Green500 list and eight of the top 20 spots, with RIKEN's Shoubu system taking first place as the first Top500 supercomputer to exceed 7 gigaflops per watt. Shoubu has a theoretical peak performance of 842.96 teraflops and a measured LINPACK of 412.67 teraflops, which puts it in 160th place on the latest Top500 list. In second and third place on the Green500 are two machines from the High Energy Accelerator Research Organization--Suiren Blue with 6.84 gigaflops per watt, and Suiren with 6.22 gigaflops per watt. The top three systems were built with PEZY's second-generation 1,024-core custom MIMD processor and Exascalar's submersion liquid-cooling technology. Suiren did not make the Green500 last year, but this year achieved higher performance and energy efficiency without any apparent hardware modifications. Suiren's LINPACK rose from 178.1 to 206.6 teraflops, and its total power consumption fell from 37.83 kW to 32.59 kW. Virginia Polytechnic Institute and State University professor Wu Feng notes the current Green500 list suggests the supercomputing community is on track to deliver a 20- to 40-megawatt exascale system by 2022. He says if researchers were still aiming for an exascale machine for the original 2018 time frame, it would be a supercomputer on the order of 150 megawatts.


Erase Obstructions From Photos With a Click
Technology Review (08/04/15) Rachel Metz

Researchers at Google and the Massachusetts Institute of Technology (MIT) have developed an algorithm to separate an image's foreground from its background in order to remove unwanted obstructions. The algorithm detects differences between the foreground and background within a sequence of photos a person has taken while moving their smartphone slightly. Google's Michael Rubinstein says the principle behind the algorithm is the phenomenon of motion parallax, in which objects closer to a person appear to move faster than those farther away. Because one of the images in the scene--the obstruction--is closer to the camera than the other, what the user is trying to take a photo of will move differently. The researchers will present their work at SIGGRAPH 2015, which takes place next week in Los Angeles. MIT graduate student Tianfan Xue, lead author of the report, says in addition to reflections on windows and chain-link fences, the algorithm can correct for several obstructions on windows, including raindrops or dirt. Xue notes it also works on other reflecting surfaces. However, the algorithm currently does not work with images of subjects in motion, in low light conditions, or on multiple obstructions in an image.


Children Beating Up Robot Inspires New Escape Maneuver System
IEEE Spectrum (08/06/15) Kate Darling

An experiment by Japanese researchers to see how mall patrons would react to a social robot turned up a surprising finding: the not infrequent hostility of children toward the robot. Researchers from ATR Intelligent Robotics and Communication Laboratories, Osaka University, Ryukoku University, and Tokai University patrolled a mall in Osaka with a remotely operated Robovie 2 robot. When someone obstructed the robots path, it would politely ask them to step aside, and would move away if they failed to do so. Although adults frequently complied with the robot's requests, the researchers found children, especially if they were in groups and there were no adults nearby, would not. Some children actively tried to frustrate the robot's movements or even attacked the robot, punching and kicking it or throwing objects at it. To counter this hostility, the researchers developed an abuse-evading algorithm that would model the likelihood an individual would become hostile. Children (anyone shorter than 4 feet 6 inches) were judged to be more likely to be hostile, especially if in unsupervised groups. If it encountered such a situation, the robot would try to move toward a taller person or a more crowded area. The research was presented at the 2015 10th ACM/IEEE International Conference on Human-Robot Interaction in Portland, OR.


A Supercomputer in the Palm of Your Hand
UMass Dartmouth (08/05/15) Joseph Sullivan

Researchers at the University of Massachusetts Dartmouth are examining the possibility of using smartphone processors as energy-efficient alternatives to current supercomputer components. The team led by professor Gaurav Khanna, associate director of the Center for Scientific Computing and Visualization Research, has a background in finding creative alternatives to standard supercomputers. Khanna and his team were among the first to see the potential of gaming consoles, in particular the Playstation 3, and graphics cards as supercomputer alternatives. Among their previous projects is a supercomputer composed of Playstation 3s stored in a refrigerated shipping container that has the processing power equivalent of nearly 3,000 typical laptop or desktop processors. At the urging of Qualcomm founder Irwin Jacobs, Khanna and his team have been testing the company's Snapdragon smartphone processors to see if they could be used as an alternative to traditional supercomputer components. Their tests so far have shown the chips use 30 times less electricity to generate the same performance as typical supercomputer servers. However, the research is still in the early stages. Khanna's team now is testing the chips to see if they can run some of the high-end scientific programs used on supercomputers.


Great Innovative Idea--Emerging Architectures for Global System Science
CCC Blog (08/05/15) Helen Wright

Michela Milano from the University of Bologna in Italy and Pascal Van Hentenryck from the University of Michigan want to look more holistically at global systems such as logistics and supply chains, health services, energy networks, financial markets, computer networks, and cities. Milano and Van Hentenryck believe now is the time to pay more attention to interactions between complex infrastructures, man-made processes, natural phenomena, and human behavior. They note relevant data sets of unprecedented scale and accuracy are now available, along with improved optimization technology and machine learning. Milano and Van Hentenryck's paper, "Emerging Architectures for Global System Science," was one of the winners at the Computing Community Consortium-sponsored Blue Sky Ideas Conference Track at the 29th Association for the Advancement of Artificial Intelligence Conference on Artificial Intelligence. Milano and Van Hentenryck say modeling and optimizing more globally could improve social welfare across the systems. "This may mean responding better to disasters, saving and producing energy more effectively, improving the quality of life in our cities and healthcare delivery, or making a supply chain more efficient and resilient," the researchers say.


MIT Camera Culture Group Develops the 'eyeSelfie' to Help Monitor Eye Health
Boston Globe (08/05/15) Vijee Venkatraman

Researchers at the Camera Culture Group, headed by Ramesh Raskar at the Massachusetts Institute of Technology (MIT) Media Lab, have developed eyeSelfie, a low-cost, handheld device for taking images of the retina, optic nerve, and vasculature. "The eyeSelfie makes retinal self-imaging possible for the first time," says Tristan Swedish, an MIT graduate student who will be demonstrating the prototype at SIGGRAPH 2015, which takes place next week in Los Angeles. The eyeSelfie, which works like a point-and-shoot camera, helps users align their gaze to enable them to take clear retinal selfies. As users peer into the interactive device, they get visual cues in the form of small red lights. Initially, they see four specks of light laid out like the tips of a diamond; if they bring the eyeSelfie closer, they see another set of dots that appear like a diamond within a diamond. When everything is aligned correctly, a ninth dot appears at the center, which is the user's cue to press the click button and take a snapshot of the interior of the eye. Experts can then evaluate the retinal selfie. In a study, the effectiveness of the device was tested on 10 volunteers, most of whom were able to successfully use the imaging technique within 10 minutes, according to the researchers.


Stanford Researchers Unveil Virtual Reality Headset That Reduces Eye Fatigue, Nausea
Stanford Report (08/03/15) Vignesh Ramachandran

New virtual reality headset technology developed at Stanford University provides a more natural viewing experience. With current "flat" stereoscopic virtual reality headsets, each eye sees only one image and the depth of field is also limited. However, Stanford professor Gordon Wetzstein says in the real world, people see slightly different perspectives of the same three-dimensional scene at different positions of their eyes' pupil. Wetzstein also notes people focus on different depths, and the disconnect between what people see and feel can cause motion sickness. The Stanford team used light-field technology to create a sort of hologram of multiple, slightly different perspectives for each eye, enabling users to freely move their focus and experience depth in the virtual scene. "You have a virtual window, which ideally looks the same as the real world, whereas today you basically have a [two-dimensional] screen in front of your eye," Wetzstein says. The prototype headset, made with off-the-shelf parts, incorporates two stacked, transparent liquid-crystal displays with a spacer. "Virtual reality gives us a new way of communicating among people, of telling stories, of experiencing all kinds of things remotely or closely," Wetzstein says. "It's going to change communication between people on a fundamental level."


Creating an Avatar From a 3D Selfie
Swiss Federal Institute of Technology in Lausanne (08/05/15)

Smartphone users will be able to generate a three-dimensional duplicate of themselves by using a new process developed by researchers at the Swiss Federal Institute of Technology in Lausanne (EPFL). Alexandru Ichim from EPFL's Computer Graphics and Geometry Laboratory and colleagues say they have condensed an expensive and complex Hollywood studio process into an application for use on smartphone cameras. "We wanted the process to be fast and easy: all you have to do is take a video of yourself and then snap a few more shots to get facial expressions, and our algorithm does the rest," Ichim says. He notes users will be able to display their digital double on a screen and animate it in real time with a video camera that follows their movements. Although creating an avatar is simple for users, developing the underlying algorithms to replace studio conditions and generate animation was challenging. Changes in the light, blurry shots without a tripod, and limited picture quality depending on the smartphone's camera were some of the factors the researchers had to account for to achieve a good result. Ichim's team plans to continue to refine the process to recreate facial expressions, and the process could potentially be used for gaming, virtual reality, films, videoconferencing, and avatar therapy in the near future.


Small Tilt in Magnets Makes Them Viable Memory Chips
UC Berkeley NewsCenter (08/03/15) Sarah Yang

University of California, Berkeley researchers have discovered a new way to switch the polarization of nanomagnets, a breakthrough they say could produce a way for high-density storage to move from hard disks onto integrated circuits. The researchers say the advance could lead to computers that turn on in an instant, operate with far greater speed, and use significantly less power. They found tilting magnets slightly makes them easy to switch without an external magnetic field, an approach that could lead to a memory system that can be packed onto a microprocessor. "To reduce the power draw and increase the speed, we want to be able to manufacture a computer chip that includes memory so that it is close to the computational action," says Berkeley professor Sayeef Salahuddin. In past studies, the researchers found directing electrical current through tantalum creates polarity in magnets without an external magnetic field. However, packing a sufficient number of nanomagnets onto a chip proved difficult, because the vertical orientation negated the switching effects of tantalum. "We found that by tilting the magnet--just 2 degrees was enough--you get all the benefits of a high-density magnetic switch without the need for an external magnetic field," Salahuddin says.


Code 'Transplant" Could Revolutionize Programming
Wired.co.uk (07/30/15) James Temperton

Researchers at University College London (UCL) have developed MuScalpel, a software tool they say is capable of automatically isolating the code of a feature in one program and "transplanting" it into another program. Research team leader Mark Harman says that like an organ transplant, a code transplant has a chance of being "rejected" by the new "host." However, because the system is automated, it can retry the transplant, hundreds or thousands of times if necessary, until it gets it right. To demonstrate the new system, Harman's team used MuScalpel to transplant the H.264 video codec from x264 into VLC media player. The automated system was able to complete the task in 26 hours, whereas VLC's manual addition of the codec took 20 days. The UCL researchers believe MuScalpel could be used to transplant many different features between programs, including everything from automatic save features and social media integration to video chat and spellcheckers. The tool currently only works on the C programming language, but the researchers say it could be adapted to work with, and even between, other languages. Harman believes the new tool could revolutionize programming, freeing programmers from much of the drudge work of developing software.


Cyber-Defense and Forensic Tool Turns 20
National Science Foundation (08/04/15) Aaron Dubrow

Users and cybersecurity engineers celebrated the open source cybersecurity defense/forensic software Bro's landmark achievements as a real-world tool as part of its 20th anniversary this week. Bro's success owes a lot to long-term backing by the U.S. National Science Foundation (NSF) and early adoption by the Lawrence Berkeley National Laboratory (LBNL), which saw its practical applications almost immediately, given that LBNL and other research centers are under near-constant cyberattack. Bro not only supports a network monitoring and traffic analysis architecture, it also enables security experts to perform forensics so they can examine attack patterns, evaluate the harm, and design better countermeasures. "For two decades, Bro has been the cornerstone of cybersecurity at Lawrence Berkeley National Lab, providing the visibility and flexibility to secure our unique network," says LBNL chief information officer Rosio Alvarez. Bro differentiates itself from commercial cybersecurity solutions via its programmability, flexibility, and ultra-high speed. In addition, customization and security community crowdsourcing enable operators to implement a programming script that tracks vulnerabilities to their source, visualizes the network impact, and facilitates rapid response. NSF's role in Bro's development included awarding grants to enhance the software, providing documentation, improving packaging, and supporting developers outside of academia.


Teach Your Robot to Do the Dishes
Technology Review (08/05/15) Mark Harris

Scientists at the University of Wisconsin-Madison (UWM) have developed a robot that can help people wash the dishes. The team taught a Kinova Mico robot arm to mimic a human handing plates from a drying rack to another person to stack on shelves after watching a handful of examples, using a Kinect sensor to track the speed and position of their arms. The algorithm predicted when a user was ready for the next dish with an accuracy of more than 90 percent. The team then programmed the robot to respond with three different strategies: work as fast as it could, sometimes wait until the receiver was completely finished with a dish before passing the next plate, and adaptively tracking humans by slowing and pausing. Users preferred working with the intuitive system over others and said they valued the user experience. "There's a trade-off between team performance and user experience," says UWM professor Bilge Mutlu. "Users want to interact with robots at their own pace, as opposed to maximizing efficiency." He believes robots will eventually help unload groceries, hand human workers parts for assembly, and guide patients through physical rehab exercises.


Intelligent Robots Don't Need to Be Conscious to Turn Against Us
Business Insider (08/05/15) Guia Marie Del Prado

In an interview, University of California, Berkeley professor Stuart Russell, founder of the university's Center for Intelligent Systems, says artificial intelligence (AI) does not have to achieve consciousness to be threatening. "When people talk about the singularity, when people talk about superintelligent AI, they're not talking about sentience or consciousness," he notes. "They're talking about superhuman ability to make high-quality decisions." Russell, who was awarded the ACM Karl V. Karlstrom Outstanding Educator Award in 2005, says most AI researchers are not focused on creating a conscious machine, but on improving systems' ability to perceive, comprehend language, and operate in the physical world. The current focus of machine design involves giving devices precisely defined goals, but Russell says the danger lies in the machine diverging from desired behavior in the pursuit of such goals. "The robot is not going to want to be switched off because you've given it a goal to achieve and being switched off is a way of failing--so it will do its best not to be switched off," he says. Russell also observes most AI progress, as embodied by deep learning, is concerned with gaining a greater mathematical understanding of tasks. He says the mathematical definition of decision-making or perception tasks enables the conception of methods for accomplishing those tasks extremely well, and those methods do not have to be identical to how humans execute them.


Abstract News © Copyright 2015 INFORMATION, INC.
Powered by Information, Inc.


To submit feedback about ACM TechNews, contact: technews@hq.acm.org
Current ACM Members: Unsubscribe/Change your email subscription by logging in at myACM.
Non-Members: Unsubscribe