Association for Computing Machinery
Welcome to the November 16, 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).


9 Research Projects That Could Transform the Enterprise
InfoWorld (11/16/15) Peter Wayner

Several university research labs are engaged in projects that could potentially transform computing, including Stanford University's DeepDive initiative to mine unstructured text for correlations by parsing natural language into streams of entities and then seeking statistical connections among them. Another project is Johns Hopkins University's ZeroCoin proposal to ensure the privacy and security of cryptocurrency transactions via a zero-knowledge proof. Meanwhile, Brown University's Burlap project combines brute-force searching and statistic exploration to enable algorithms to search node networks to find the best solutions to problems. Halide from the Massachusetts Institute of Technology (MIT) is an image-processing computer language designed to abstract myriad tasks for digital photography enhancement. MIT also is behind Visual Microphone, a software project that uses images to detect minuscule object movements, and Drake, a series of packages to ease development of software code governing robots and drones. Meanwhile, the University of Washington's SpiroSmart software can enable smartphones to diagnose lung health by listening to a patient's breathing and measuring echoes and reverberations, and the Vienna University of Economics and Business is hosting the R project for extracting statistical insights from big data that prove or disprove hypotheses. Finally, many universities are using GitHub repositories to open source and share their courses.

'Shrinking Bull's-Eye' Algorithm Speeds Up Complex Modeling From Days to Hours
MIT News (11/16/15) Jennifer Chu

Massachusetts Institute of Technology (MIT) researchers have devised an algorithm that vastly reduces the computation of virtually any computational model. Over several runs of a model, and in tandem with relevant data points, the algorithm incrementally narrows in on a probability distribution of values for each unknown parameter. The researchers used this method to arrive at the same answer as a classic computational strategy for relatively complicated models, speeding up the time 200-fold. Computer scientists usually use Markov chain Monte Carlo analysis when working with complex models involving multiple unknown parameters, the catch being that this technique could take days to weeks to produce an answer. "What our algorithm does is short-circuits this model and puts in an approximate model," says MIT professor Youssef Marzouk. "It may be orders of magnitude cheaper to evaluate." At the outset of the analysis, the algorithm outlines large, vague bull's-eyes over the probability distribution's entire scope. After successive runs with either the model or the data, the bull's-eyes progressively contract, focusing on the spaces, or values, most likely to represent the unknown parameter.

Supercomputer Leaders Come Together on New Open Source Framework
ZDNet (11/12/15) Steven J. Vaughn-Nichols

The Linux Foundation and a consortium of high-performance computing (HPC) leaders last week announced the creation of the OpenHPC Collaborative Project, which has the goal of creating a new open source framework to support the world's most sophisticated HPC environments. Members of the new consortium include several national laboratories and academic supercomputing facilities, as well as Intel, Hewlett-Packard, Dell, and Lenovo. Currently, 97 percent of the world's fastest supercomputers run Linux. However, as systems become faster and more powerful, a new and dedicated solution for HPC increasingly is warranted. "The use of open source software is central to HPC, but lack of a unified community across key stakeholders...has caused duplication of effort and has increased the barrier to entry," says Linux Foundation executive director Jim Zemlin. "OpenHPC will provide a neutral forum to develop one open source framework that satisfies a diverse set of cluster environment use-cases." The new initiative also is being driven by increased interest in HPC among the private sector as an aid to big data analytics. Analysts note this is especially true in the world of finance, which has significantly increased its supercomputing efforts in recent years.

Georgia Tech Trains Watson AI to 'Chat,' Spark More Creativity in Humans
Georgia Tech News Center (11/12/15) Joshua Preston

Georgia Institute of Technology (Georgia Tech) researchers are using IBM's Watson to advance how computers could help humans creatively solve problems. They trained Watson using 1,200 question-answer pairs, which enabled them to "chat" with Watson about design challenges in areas such as engineering, architecture, systems, and computing. The researchers then posed questions to Watson to see what it had learned, and it was able to answer the questions and guide students through the task of examining a wide volume of research that may fall outside their expertise. The researchers say their approach, called GT-Watson Plus, could assist professionals in a variety of fields by enabling them to ask questions and receive answers as quickly as in natural conversation to help with problem-solving. Watson's ability to retrieve natural-language information would enable a novice to quickly learn about complex topics and better determine whether their idea or hypothesis is worth pursuing, according to the researchers. GT-Watson Plus also prompts users with alternate ways to ask questions for better results, which are then packaged in an intuitive presentation that enables the average person to navigate results more easily on a given topic. "We were able to add more semantic and contextual meaning to Watson to give some notion of a conversation with the [artificial intelligence]," says Georgia Tech professor Ashok Goel.

Crash Test Simulations Expose Real Risks
National Science Foundation (11/12/15) Aaron Dubrow

Researchers at the Virginia Polytechnic Institute-Wake Forest University Center for Injury Biomechanics are developing computer models of vehicle crashes to provide more sophisticated information on how to improve restraints and other safety systems. The models also help researchers simulate the effects of thousands of variables that would be far too slow to test in a physical crash test. The researchers are working with the Crash Injury Research and Engineering Network, which has created a database of real-world vehicle crashes for researchers to test with computer models. The researchers used the U.S. National Science Foundation-supported Blacklight supercomputer at the Pittsburgh Supercomputing Center and the DEAC Cluster at Wake Forest to run thousands of simulations taken from hundreds of cases. They also worked with members of the Extreme Science and Engineering Discovery Environment Extended Collaborative Support Service team, who helped set up the cyberinfrastructure and workflows needed to run the simulations. The researchers showed simulations can reproduce real-world injury patterns and predict details crash-test dummies cannot provide. "By studying a variety of potential occupant positions, we can understand important factors that lead to more severe injuries and potentially mitigate these injuries with advanced safety systems to protect occupants in more dangerous positions," the researchers say.

Track Your Heart With Your Phone, Even If Your Phone's in Your Bag
Technology Review (11/12/15) Rachel Metz

Massachusetts Institute of Technology (MIT) researchers are working on a project called BioPhone, which derives a person's heart and breathing rates from a smartphone's accelerometer. MIT Media Lab's Javier Hernandez Rivera, the lead author of a paper on the project, says the initiative is meant to capture this data during moments when a person is not moving much. He says the data could be used to determine if a person is stressed and provide ways to cope, such as breathing exercises or asking a loved one to call. Researchers tested BioPhone by having 12 study participants sit, stand, and lie down before and after pedaling on an exercise bike with an Android smartphone in their pocket to capture data from the phone's accelerometer. During the tests they wore sensors for capturing heart and breathing rates to enable the data gathered from the smartphones to be compared with measurements already known to be reliable. On average, the heart rates estimated by analyzing the smartphone data were off by slightly more than one beat per minute, and breathing rate estimations were off by about a quarter of a breath per minute. Hernandez says there are still challenges to overcome, such as how to measure heart and breathing rates reliably when the smartphone is in different spots.

Machine Learning Could Solve Riddles of Galaxy Formation
University of Illinois News Bureau (11/11/15) Austin Keating

University of Illinois researchers have developed a machine-learning algorithm that maps the relationship between dark matter halos and their normal matter counterparts in hydrodynamical galaxy simulations. The researchers say the new machine-learning simulation system is a necessary first step to developing more accurate and relevant insights into the formation of the universe. Machine-learning algorithms reduce computing time by approximating the properties a researcher wants to examine using an algorithm that has been trained on a hydrodynamical simulation. The researchers used the algorithm to show the predicted distribution of galaxies and their properties is almost exactly correct, despite the fact that the placement of dark matter and particles might be slightly off. The researchers also demonstrated that machine-learning technology was able to recreate a distribution of galaxies similar to those produced by semianalytical models. In addition, they compared the machine-learning method to hydrodynamical simulations, and again found the machine-learning paradigm performed well. The researchers note the new method could be used to quickly create simulated maps of galaxies to compare to observations.

Web Tool Helps People Visualize, Make Sense of Large Complex Datasets
Carnegie Mellon News (PA) (11/10/15) Byron Spice

Current datasets often grow so large and complex that automated methods appear to be the only way to gain knowledge from them. A new Web-based tool being developed at Carnegie Mellon University (CMU) offers the option to retain human judgment and intuition in analysis. Called Explorable Visual Analytics (EVA), the tool uses a computer architecture that enables analysts to explore raw data through dynamic visualizations with minimal time delay. The goal is to enable users to make sense of "high-dimensional" data, or data with many parameters. The bulk of the data can reside in an external network, while EVA downloads to the user's computer only the portion being analyzed. The data-processing pipeline includes pre-processing and caching of data on servers, compressing data to limit the use of communications bandwidth, and caching data on the client computer for better responsiveness. "We are able to give users the illusion they are working with all of a massive dataset while actually sending only a small proportion of the data to the client," says CMU researcher Amir Yahyavi. He says EVA's rapid response enables users to quickly explore different parameters and examine them using the most appropriate types of graphics. EVA also helps in communicating findings from the data so users can share their conclusions and the process used to reach them.

New Algorithm Identifies Lockstep Reputation Attacks
The Stack (UK) (11/10/15) Martin Anderson

Researchers from the University of Sao Paolo say they have developed an algorithm that can identify fake online product reviews and even identify which are likely to have been originated by a product or company's competitors. Fake reviews have become a significant problem for many of the websites that aggregate or rely on user reviews. TripAdvisor, for example, has faced intermittent legal challenges related to fake reviews on its platform, including one case in Italy in 2011 that led to the company being fined. Amazon also struggles with fake reviews, and the company recently filed lawsuits against more than 1,000 users who had submitted fake reviews. The Sao Paolo researchers' Online-Recommendation Fraud ExcLuder (ORFEL) algorithm is focused on identifying lockstep behavior that can indicate a group of users are coordinating their efforts and interactions with a given product. ORFEL relies on a vertex-centric algorithm that can individuate lockstep traces in Web-scale graphs using parallel processing. The researchers note ORFEL's more general approach enables it to identify not only negative smear campaigns, but also efforts by fake reviewers to boost a product or service. They say ORFEL can identify fake reviews with 95-percent accuracy.

Positive Emotions Are More Contagious Than Negative Ones on Twitter
USC News (11/09/15) Robert Perkins

Researchers have used an algorithm designed to determine the emotional content of Tweets to study how emotions propagate through Twitter. University of Southern California's Emilio Ferrara and Indiana University's Zeyao Yang analyzed about 3,800 randomly chosen Twitter users, using an algorithm that ranked their Tweets as being either positive, negative, or neutral. They then compared the sentiments of users' tweets to the ratio of sentiments of all the tweets that appeared in those users' feeds during the hour before. Higher-than-average numbers of positive tweets in a user's feed were associated with that user making more positive tweets. The same was true of negative tweets, where higher numbers of negative tweets in their feeds prompted users to tweet more negatively. However, the researchers found this effect was much stronger for positive tweets. About 20 percent of the Twitter users deemed to be highly susceptible to what the researchers called "emotional contagion" were four times more likely to be affected by positive tweets than negative tweets. Even those least affected by emotional contagion were still slightly less than twice as likely to be more affected by positive than negative tweets.

Disney Software Makes It Easy to Design and Print Custom Walking Robots
IEEE Spectrum (11/09/15) Evan Ackerman

A joint project between ETH Zurich, Carnegie Mellon University, and Disney Research has yielded an interactive design system that enables hobbyists to create custom walking robots that can be printed in three dimensions (3D). Each robot starts out with an initial skeleton in which virtual motors connect the bones at each joint position. Walking is facilitated by adjusting which legs are on the ground when, while keeping the robot from toppling by making sure the green ball representing its center of mass remains within the red box representing the stability polygon created by whichever legs are in contact with the ground. The software optimizes the motor values to produce dynamically stable motions that can then be previewed in a physics-based model. The final stage is to generate 3D geometry for all of the robot's body parts, including motor connectors. The software accounts for what kind of 3D printer and constituent materials the user is employing. The software's developers constructed a variety of robots from scratch, and they say real-world testing demonstrated "good agreement between the overall motions of our physical prototypes and the behavior predicted in simulation." The research was presented at the recent ACM SIGGRAPH Asia 2015 conference in Kobe, Japan.

Artificial Neuronal Network Learns to Use Human Language
Science 2.0 (11/11/15)

Researchers from the University of Sassari and the University of Plymouth have developed a cognitive model comprising 2 million interconnected artificial neurons called the Artificial Neural Network with Adaptive Behavior Exploited for Language Learning (ANNABELL), which is able to learn to communicate using human language. The model can learn human language starting from a blank state only via communication with a human interlocutor. ANNABELL is a cognitive architecture consisting entirely of interconnected artificial neurons, and does not have pre-coded language knowledge. It relies on synaptic plasticity and neural gating. Synaptic plasticity is the ability of the connection between two neurons to increase its efficiency when the two neurons are often active simultaneously, or nearly simultaneously. Neural gating mechanisms are based on the properties of bistable neurons to behave as switches that can be turned "on" or "off" by a control signal coming from other neurons. The cognitive model has been validated using a database of about 1,500 input sentences, based on literature on early language development, and has responded by generating a total of about 500 output sentences containing nouns, verbs, adjectives, pronouns, and other word classes. The research demonstrates the model's ability to express a wide range of capabilities in human-language processing.

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

To submit feedback about ACM TechNews, contact: [email protected]
Current ACM Members: Unsubscribe/Change your email subscription by logging in at myACM.
Non-Members: Unsubscribe