Welcome to the July 22, 2016 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
Report Addresses the Perils of Dark Silicon
HPC Wire (07/21/16) Tiffany Trader
The authors of a report on dark silicon--the lost processing potential resulting from inhibited full central-processing unit usage because of thermal limitations--warn with future technology nodes, it will be "infeasible to operate all on-chip components at full performance at the same time due to the thermal constraints (peak temperature, spatial and temporal thermal gradients, etc.)." High-performance computing has fewer restrictions since heat is stripped with cooling techniques, but with exascale systems coming soon, power grid limitations and energy costs pose serious issues. The study estimates the five-year energy costs of today's largest systems are about equal to the purchase price. According to the study authors, many of the fastest supercomputers do not exercise their full-power capacity past setup and benchmarking. Mission-critical simulation codes seldom top 60 percent of allocated power, and the remaining percentage is dark. The researchers propose hardware overprovisioning as a realistic solution to this challenge in the exascale timeframe. "This solution requires buying more compute resources than can be executed at maximum power draw simultaneously," they note. To achieve this feasibly, the researchers say power management must be engineered "as a first-class resource at the level of the scheduler, the run-time system, and on individual nodes."
How the World's Most Powerful Supercomputer Inched Toward the Exascale
IEEE Spectrum (07/20/16) Rachel Courtland
China's Sunway TaihuLight in June topped the Top500 list as the world's most powerful supercomputer, capable of performing the Linpack Benchmark at a rate of 93 petaflops while consuming 2.4 megawatts less power than the previous record-holding system. Such efficiency upgrades are required in the push toward exascale operation, and TaihuLight uses "lightweight" microprocessors that process data slower but produce less heat, allowing them to be packed more densely. Although the system can theoretically crunch numbers at a rate of 125 petaflops, it reaches 74 percent of this peak theoretical capacity when running Linpack. However, according to the High Performance Conjugate Gradients benchmark, TaihuLight uses only 0.3 percent of its theoretical peak abilities. "They produced a processor that can deliver high-arithmetic performance but is very weak in terms of data movement," says University of Tennessee professor Jack Dongarra, one of the organizers of the Top500. He notes the TaihuLight team has created applications that leverage the architecture, and three apps designed to run on the machine were finalists for this year's ACM Gordon Bell Prize. University of Manchester professor John Goodacre says TaihuLight's design reduces the energy cost of shuttling data back and forth. "I think what they've done is build a machine that changes some of the design rules that people have assumed are part of the requirements" for migrating to the exascale, he says.
Baidu Uses Millions of Users' Location Data to Make Predictions
New Scientist (07/20/16) Hal Hodson
The Big Data Lab at Chinese search engine Baidu has compiled billions of location records from its users to build an employment and consumption index, based on the flux of people around commercial and industrial zones. The Lab labeled thousands of offices, commercial centers, and industrial zones across China, and tracked how many people visited those areas. The data, spanning from 2014 to the middle of 2016, reflects shuttered businesses and startup successes via changes in foot traffic. Baidu's employment index shows manufacturing employment has declined by about 10 percent since 2014, while employment in the technology industry has grown. A similar index was created to reflect Chinese consumer activity. Baidu has issued a prediction for Apple's Chinese revenue in the second quarter based on data gathered about traffic to Apple Stores, predicting a 20-percent drop in Apple revenue. "To the best of our knowledge, we are the first to measure the second-largest economy by mining such unprecedentedly large-scale and fine granular spatial-temporal data," the Baidu researchers say.
CCC Computing Research Symposium--Life Long Learning (Education and Workforce)
CCC Blog (07/20/16) Vasant G. Honavar
The automation of blue-collar and white-collar jobs via advances in robotics and artificial intelligence (AI) raises questions about whether those innovations will help create new types of occupations, how individuals can prepare themselves to succeed in a rapidly changing workforce, and other issues discussed at the Computing Community Consortium's Symposium on Computing Research: Addressing National Priorities and Societal Needs. "Communications of the ACM" editor-in-chief and Rice University professor Moshe Vardi discussed AI's moral imperatives. "As computing professionals, we...have a moral imperative to acknowledge the adverse societal consequences of the technology we develop and to engage with social scientists to find ways to address these consequences," he said. University of Washington professor Zoran Popovic's presentation on technology-enabled strategies for developing human expertise highlighted the prospects of human-machine systems to ready an agile workforce for swift and disruptive technological change. ACT chief innovation officer Miguel Encarnacao discussed how workforce development is challenged by advances in computing, AI, and robotics. He emphasized learning analytics and personalized learning/teaching platforms as tools for tailor-made content and content delivery so individuals can gain the knowledge and skills to succeed and prosper.
Program Gets Women Coding With NASA Data
FedScoop (07/20/16) Samantha Ehlinger
The U.S. National Aeronautics and Space Administration (NASA) recently named 49 participants to its 2016 Datanauts program, which aims to give more women the opportunity to use its data and code with it. The participants, who all have varying levels of programming and data science experience, will be presented with advanced, month-long challenges that will build their coding skills. NASA came up with the idea for the program after realizing participants in its International Space Apps Challenge were overwhelmingly male. Some of the challenges the participants will face over the course of the six-month program are from Space Apps, while others are designed specifically for Datanauts. The sophomore class consists of mostly adults, but includes some university students and one high school student, from across the U.S. The founding class was all female, but there are five men in the group this year. The inaugural Datanauts class advised the team on how to design the program, says Beth Beck in the Office of the CIO at NASA. The Datanauts also will participate in a one-day boot camp to help prepare them for Space Apps and will have the opportunity to use specially designed toolkits to plan events in their own communities.
Columbia Engineering Researchers Use Acoustic Voxels to Embed Sound With Data
Columbia University (07/18/16) Holly Evarts
A project by researchers at Columbia University, the Massachusetts Institute of Technology, and Disney Research has yielded a computational technique to control sound waves by inversely designing acoustic filters that can fit within an arbitrary three-dimensional (3D) shape while supporting target sound-filtering properties. The team designed small, hollow, cube-shaped acoustic voxels through which sound enters and exits, and which can be connected into an infinitely adjustable, complex arrangement. The voxels' internal chambers enable filtering modification, so changing their number and size or how they link alters the resulting acoustics. "Our algorithm enables new designs of noise mufflers, hearing aids, wind instruments, and more--we can now make them in any shape we want, even a 3D-printed toy hippopotamus that sounds like a trumpet," says Columbia professor Changxi Zheng. He also notes his team has proposed a novel filter application in which the voxels are employed as acoustic tags, unique to each 3D-printed piece, and encoded with information. The researchers recorded the sound produced by objects with different voxel structures and used an iPhone app they developed to accurately identify each object. Zheng notes if manufacturers can implant identifying information directly within an object, they will be spared the time, effort, and cost of individually labeling parts.
Imaging Software Predicts How You Look With Different Hair Styles, Colors, Appearances
UW Today (07/21/16) Jennifer Langston
Computer vision research at the University of Washington (UW) has yielded a personalized image search engine that enables users to predict their appearance with a different hairstyle or color, or as they would look in a different time period, age, or country. "This is a way to try on different looks or personas without actually changing your physical appearance," says UW professor Ira Kemelmacher-Shlizerman. "While imagining what you'd look like with a new hairstyle is mind-blowing, it also lets you experiment with creative imaginative scenarios." The method, to be presented next week at the ACM SIGGRAPH 2016 conference in Anaheim, CA, involves the user uploading an input photo, and then entering a search term. The software's algorithms sift through online photo collections for similar images in that category, mapping the person's face onto the results. The system builds on earlier UW work in facial processing, recognition, three-dimensional reconstruction, and age progression, integrating those algorithms to generate the blended images. The new software also can be used to show how a missing child or fugitive might appear in different disguises or in later years. Kemelmacher-Shlizerman's team previously developed automated age-progression software that concentrated solely on a person's face, while the new system adds varied hairstyle options and other contextual components.
Character Animation Technique Produces Realistic-Looking Bends at Joints
Phys.org (07/18/16) Jennifer Liu
Disney researchers have developed a method to pre-compute an optimized center of rotation for each vertex in a character model, and those centers of rotation could be the basis for calculating how the skin around each joint is deformed as it is bent. "The pre-computation enabled us to significantly reduce the joint distortions that often plague these animations, preserving the volume of the skin surface around the joint," says Disney Research's Jessica Hodgins. In addition, she says this method can be dropped into the standard animation pipeline. Currently, two skinning methods, called linear blend skinning (LBS) and dual quaternion skinning (DQS), are widely used in computer game engines, virtual reality engines, and in three-dimensional animation software. However, researchers say both methods have difficulty with certain poses. Disney postdoctoral researcher Binh Huy Le says pre-computing the centers of rotation solves this issue by improving the ability to properly weight the influence of each bone in the joint on the skin deformation. Le notes the new method minimizes or eliminates the volume losses associated with LBS and the bulging associated with DQS. The researchers will present the skeletal skinning method next week at the ACM SIGGRAPH 2016 conference in Anaheim, CA.
Faster Prediction of Wireless Downtime
King Abdullah University of Science and Technology (07/18/16)
Researchers at King Abdullah University of Science and Technology (KAUST) have used an importance sampling technique to simulate rare events for the problem of wireless outage capacity. The outage capacity measures the percentage of time the communication system undergoes an outage, which is normally about one second per million or more. "There are no efficient analytical solutions to this problem, and to simulate this situation using conventional simulation methods might take more than a billion simulation runs," says KAUST researcher Raul Tempone. In order to solve this problem, the researchers used importance sampling, an approach through which a clever problem transformation makes it possible to sample more frequently from the event of interest, turning rare events in the original problem into non-rare events in the transformed problem. The importance sampling approach is suitable for a wide range of challenging network scenarios. "Despite continuous advances in the concept of importance sampling in the field of rare events simulations, its popularity among researchers in the field of wireless communication systems is still quite limited," Tempone says. "Our work is the first to bridge the gap between the framework of rare event algorithms and the evaluation of outage capacity for wireless communication systems."
Robot Therapist Hits the Spot With Athletes
Nanyang Technological University (Singapore) (07/18/16) Lester Kok
A robotic massage therapist is now seeing patients at the Singapore Sports Hub, treating injured athletes using acupoint therapy and Traditional Chinese Medicine (TCM) massage techniques. Expert Manipulative Massage Automation (Emma) was designed by AiTreat, a startup incubated by Nanyang Technological University (NTU), to meet a growing demand in Singapore for trained therapists and consistent treatment. Emma consists of proprietary software running a robotic arm and a fully rotatable massage tip constructed by a three-dimensional (3D) printer. The robot also uses a 3D-stereoscopic camera and pressure sensors to monitor the patient's safety and comfort. Emma's diagnostic functions are then uploaded to the cloud, enabling further analysis of the patient's recovery progress. The massage functions were designed to integrate advanced sports science and TCM pain management, based on NTU graduate Albert Zhang's experience as a licensed TCM physician. "Our aim is not to replace the therapists who are skilled in sports massage and acupoint therapy, but to improve productivity by enabling one therapist to treat multiple patients with the help of our robots," Zhang says. Since patient trials began, Emma has treated 50 patients suffering from a variety of injuries.
Blockchains: Focusing on Bitcoin Misses the Real Revolution in Digital Trust
The Conversation (07/18/16) Ari Juels; Ittay Eyal
The bitcoin cryptocurrency proves the viability of the underlying blockchain technology as a tool for ensuring trust in digital records and transactions, according to Cornell University professor Ari Juels and research associate Ittay Eyal. They say the blockchain offers unmatched security and reliability, with advocates convinced blockchains will address a range of challenges such as stabilization of financial systems, identification of stateless individuals, establishing title to real estate and media, and efficient supply chain management. Juels and Eyal say the blockchain's purpose is to support a publicly available "distributed ledger" that lists only authorized transactions whose data is indelible. "This feature makes it possible to prevent account holders from reneging on transactions, even if their identities remain anonymous," the researchers note. They also point out the ledgers' decentralized architecture offers resilience to both technical and political failures. Juels and Eyal say blockchains could be augmented to also support smart contracts, or codes that control blockchain assets to ensure predictable behavior. "The transparency and accountability of a fully public ledger have many benefits, but are at odds with confidentiality," the researchers acknowledge. Guaranteeing that smart contracts rightly reflect user intent is another challenge, but Juels and Eyal think the blockchain's applications should broaden once such obstacles are surmounted.
Partially Automated Cars Provide Enough Benefits to Warrant Widespread Adoption of current safety technologies
Carnegie Mellon University (07/18/16) Sherry Stokes
Carnegie Mellon University (CMU) researchers have concluded the public could gain economic and social benefits if safety-oriented, partially automated vehicle technologies were deployed in all cars. The researchers came to this conclusion following an examination of forward collision warning, lane departure warning, and blind spot monitoring systems. These technologies can include partially autonomous braking or controls to help vehicles avoid crashes. "While there is much discussion about driverless vehicles, we have demonstrated that even with partial automation there are financial and safety benefits," says Chris T. Hendrickson, director of the Carnegie Mellon Traffic21 Institute. The researchers analyzed the benefits and costs of deploying crash-avoidance technologies in the U.S. light-duty vehicle fleet, and found they could prevent or reduce the severity of up to 1.3 million crashes a year, including 10,100 fatal wrecks. The researchers analyzed government and insurance industry data to determine if it is economically advantageous to accelerate deployment of these technologies. They say in the perfect-world scenario in which all relevant crashes are avoided with these technologies, there is an annual benefit of $202 billion, or $861 for every car. "This study creates a framework for regulatory action encouraging early deployment of partial automation technologies," Hendrickson says.
The Machine Data Challenge Cancer Researchers Face
CIO Australia (07/14/16) Byron Connolly
The application of machine learning methods to cancer research is challenging because of a lack of clean data, according to University of Texas professor and control theorist Mathukumalli Vidyasagar. "The biological processes are very noisy so whatever measurements you take, they are not very repeatable," he notes. "In other words, if we were to take the same measurement on the same tumor on the same instrument on two consecutive days, you will get somewhat different numbers." Vidyasagar's solution to this problem is to develop new machine learning algorithms on an as-needed basis. He says some algorithms have been used to help anticipate drug effectiveness on lung cancer, or to predict how much time will pass before ovarian cancer patients relapse so doctors can better schedule follow-ups. Vidyasagar says the speed of processing machine data in cancer biology makes little difference to outcomes, compared to other sectors where real-time number crunching is imperative. He says in an ideal scenario, machine learning teams should collaborate with doctors, which rarely happens. "The clinician community is a little bit unfamiliar with mathematical methods and if you use some machine learning algorithms to [produce results], they are not necessarily willing to try some of those ideas," Vidyasagar notes.
Abstract News © Copyright 2016 INFORMATION, INC.
To submit feedback about ACM TechNews, contact: [email protected]
Current ACM Members: Unsubscribe/Change your email subscription by logging in at myACM.