Welcome to the March 9, 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
Google's AI Wins First Game in Historic Match With Go Champion
Wired (03/09/17) Cade Metz
A Google computer program this week beat Go champion Lee Sedol in the first round of a five-game tournament in Seoul, South Korea. AlphaGo, developed by Google's DeepMind lab over the last two years, differs from previous Go systems by having a much greater reliance on machine learning than on a pre-set roster of maneuvers. DeepMind expedited computer Go-playing skill with two complementary machine-learning techniques: learning tasks via an analysis of vast datasets and self-administered practice of those tasks. AlphaGo mastered the game over the past five months by playing against itself using reinforcement learning technology. DeepMind CEO Demis Hassabis notes the developer team also used machine-learning methods to improve AlphaGo's time management capabilities. English-language game commentator Michael Redmond observes the program plans its moves to reinforce its weak groups, "and potentially create weak groups [for its opponent]." AlphaGo employs machine-learning techniques to narrow the scope of potentially beneficial maneuvers, but then applies a tree search to calculate the possible outcomes of those moves. The match with Sedol, which extends through next Tuesday, has generated a lot of interest among the general population, as well as within Google. "I expected it to be big," Hassabis says. "But not that big."
White House Expands Program to Spur State, Local Tech Hiring
FedScoop (03/09/16) Alex Koma
The White House announced on Wednesday the expansion of its TechHire initiative to 15 new U.S. communities to encourage state and local leaders to ramp up their efforts to recruit technology talent via private-sector partnerships. Jacob Leibenluft with the White House's National Economic Council cites U.S. Labor Department statistics that more than 500,000 information technology jobs are currently vacant across the U.S. According to Leibenluft, "too many Americans think these jobs are out of their reach, that they're only in Silicon Valley or require advanced degrees, but that's just not true." He acknowledges the TechHire initiative does not have accompanying federal funding, but says the Obama administration's "call to action" is far more valuable than any currency to prompt new commitments throughout the community. Rhode Island Gov. Gina Raimondo says since teaming with the White House six months ago, she has successfully enlisted 35 private-sector partners, which have committed to hiring workers of all backgrounds for tech jobs. She emphasizes the administration's support as essential to convincing employers to hire skilled candidates who lack "textbook, four-year degrees." Complementing TechHire are additional White House programs to bolster the tech education pipeline, including a Career and Technical Education Makeover Challenge from the U.S. Education Department; this contest will split $200,000 among 10 high schools to help develop tech-focused "makerspaces."
Quantum Mechanics Is So Weird That Scientists Need AI to Design Experiments
CNet (03/07/16) Michelle Starr
University of Vienna researchers have developed Melvin, an algorithm that helps design experiments in quantum optics. The research began with doctoral student Mario Krenn, who was trying to design a particular experiment by putting together lasers and mirrors in such a way that would lead to a specific quantum state. At one point, Krenn realized he was merely guessing, and an algorithm would be able to guess just as well as a human, but much faster. Krenn defined the goal, developed an algorithm, and let it run overnight, and in the morning the algorithm had produced a solution.txt file. The algorithm works by taking the foundation of a quantum experiment, in this case lasers and mirrors, and the quantum state desired as an outcome, and randomly runs through different setups. If a random setup results in the desired outcome, Melvin will simplify it. In addition, the algorithm can learn from experience, remembering which configurations result in which outcomes, so it can use those and develop them as needed. Krenn used Melvin on Greenberger-Horne-Zeilinger (GHZ) states, and the algorithm produced 51 experiments resulting in entangled states, including one that delivered the GHZ state.
Theoretical Foundations for Social Computing Workshop Report
CCC Blog (03/04/16) Helen Wright
The organizing committee for the Computing Community Consortium-sponsored Theoretical Foundations for Social Computing Workshop held last June have published their report on the workshop. About 25 experts in related fields were gathered at the event to talk about the potential and challenges of creating mathematical foundations for social computing. Among the social computing advancements based on mathematical research recognized by the report is the design of new systems in which hundreds or millions of individuals can work together to reach agreement on societal issues. Also lauded are the application of algorithmic research to design ethical and efficient pricing mechanisms for prediction markets, and the use of social-computing systems to help groups of people make fair-division decisions in their daily lives. Moreover, the report identifies challenges to be solved if mathematical research is to contribute significantly to social computing. Among such challenges are combining complementary mathematical and experimental research to establish relevant platforms for social computing; basing the mathematical foundations on models that better illustrate human behavior; incorporating reusable components into those models, so results can generalize to numerous systems; and instilling transparency, interpretability, and fairness within social computing algorithms and models.
AI Crossword-Solving Application Could Make Machines Better at Understanding Language
University of Cambridge (03/07/16) Sarah Collins
British, U.S., and Canadian researchers say they have developed a Web-based machine-language platform that solves crosswords better than existing commercial products by using artificial neural networks, and it could potentially improve machines' language understanding. Tests showed the system could more accurately answer single-word, short-word combinations, or sentence/phrase clues than commercial software, and also function as a "reverse dictionary" in which the user describes a concept and the platform yields possible descriptive words. The researchers "trained" the software to recognize words, phrases, and sentences using definitions in six dictionaries and Wikipedia. "We're seeing a lot more usage of deep learning, which is especially useful for language perception and speech recognition," notes the University of Cambridge's Felix Hill. He says definitions contain an important clue for helping models interpret and represent phrase and sentence meaning. "Our system can't go too far beyond the dictionary data on which it was trained, but the ways in which it can are interesting, and make it a surprisingly robust question-and-answer system--and quite good at solving crossword puzzles," Hill notes. The platform has limitations, such as an inability to infer a user's intent or wider context when it receives a query.
Big Data for Text: Next-Generation Text Understanding and Analysis
Saarland University (03/07/16)
Researchers at the Max Planck Institute for Informatics in Saarbrucken say they have developed novel text-analysis technology that significantly improves searching very large text collections using artificial intelligence. In addition, they say the technology could help authors in researching and writing texts by automatically pulling background information and suggesting links to relevant websites. Analyzing texts often requires extremely large knowledge graphs built upon freely available sources or large media portals on the Web. By applying algorithms, the texts are screened further and analyzed with linguistic tools. "Our software then assigns companies and areas of business to their corresponding categories, which allows us to gather valuable insights on how well one's own products are positioned in the market in comparison to those of the competitors," says Max Planck Institute for Informatics researcher Johannes Hoffart. He says the technology helps to map words and phrases to their correct objects in the real world, automatically resolving ambiguities. The researchers say the algorithms could be used by organizations that analyze online media or social networks to measure the degree of brand awareness for a product or the success of a marketing campaign.
Wi-Fi Breadcrumbs Reveal Pedestrian Patterns
Swiss Federal Institute of Technology in Lausanne (03/07/16)
The traces people leave as they pass Wi-Fi access points could provide key insights into the comings and goings of pedestrians. Antonin Danalet, a researcher in the Transport and Mobility Laboratory at the Swiss Federal Institute of Technology in Lausanne (EPFL), used this cost-effective approach to study the eating destinations of students on campus. He recorded more than 2 million points picked up by the campus' 789 Wi-Fi antennas over 10 days in 2012. EPFL's information technology service provided anonymized data that enabled Danalet to link the points he collected with about 2,000 individuals. Danalet merged Wi-Fi location data with map data, and was able to reconstruct routes and identify when pedestrians reached their destination, how long they spent there, and what they did. He verified the accuracy of his model partly by comparing the Wi-Fi traces to the results of surveys asking students about their movement on campus. Danalet says the approach can predict changes in what people do and where they go, and could provide clues about the use of pedestrian infrastructure at music festivals, museums, and hospitals.
Leaf Mysteries Revealed Through the Computer's Eye
Penn State News (03/07/16) A'ndrea Elyse Messer
A computer-vision method has enabled an international team of researchers to quickly classify leaves and generate vast quantities of new botanical data. The system is not programmed to exhibit a particular behavior, but instead it learns from leaf images together with category labels corresponding to family and order. The researchers say they have achieved a 72-percent accuracy rate over 19 leaf families compared to about 5 percent for random chance. The researchers provide the program with half the images already identified so it can automatically learn a dictionary of special features, which are critical to identifying leaves. The system also learns to ignore the typical problems of low image quality, insect bites, and mounting defects, and the algorithm then receives unlabeled test photos and uses its dictionary to identify them. The researchers repeated this procedure 10 times, randomly selecting the training and test images; the outcomes agreed, with only a 1-percent difference between runs. The computer generates a "heat" map directly on the leaf image, identifying and rating areas of importance for correct identification. "Variation in leaf shape and venation, whether living or fossil, is far too complex for conventional botanical terminology to capture," says Pennsylvania State University professor Peter Wilf. "Computers, on the other hand, have no such limitation."
The Supremely Intelligent Rat-Cyborg
Phys.org (03/07/16) Emilie Reas
Zhejiang University researchers recently conducted a study comparing the problem-solving abilities of rats, computers, and rat-computer "cyborgs." The researchers trained six rats over the course of a week to run a series of unique mazes. The rats' brains were implanted with microelectrodes. The rats were enticed to complete the maze by the smell of peanut butter, a sip of water, and stimulation of the pleasure centers of the brain once they solved the puzzle. After training, the researchers tested the rats on 14 new mazes, monitoring their paths, strategies, and time spent solving the mazes. The researchers also developed a maze-solving algorithm implementing left- and right-hand wall-following rules, and compared the performance of the rats to that of the algorithm. The rat cyborgs integrated the computational powers of organic and artificial-intelligence systems. The researchers compared the performance of unassisted rats, the algorithm, and the rat-cyborgs by evaluating how many steps they took, how many locations they visited, and the total time spent reaching the target. Although the cyborgs and the algorithm took about the same number of steps, the cyborgs took fewer steps than the unassisted rats, indicating more efficient problem solving. In addition, the cyborgs visited fewer locations than the algorithm or the unassisted rats, and took less time than the unassisted rats to solve the mazes.
First Code of Conduct for the Use of Virtual Reality Established
Johannes Gutenberg University of Mainz (Germany) (03/04/16)
Philosophers at the Johannes Gutenberg University of Mainz believe the use of virtual reality (VR) by researchers and the general public could present ethical issues. The technology needed to generate virtual worlds from home computers is expected to soon be widely available to the general public. Professors Michael Madary and Thomas Metzinger are particularly concerned about the unanticipated consequences of the technology on the psychological states and self-images of users. According to Madary and Metzinger, participants in VR studies have showed strong emotional reactions and behavioral changes, suggesting the technology could have an impact on their lives. They say VR creates a scenario in which the user's bodily appearance and visual environment are determined by the host of the virtual world, which raises the possibility VR will create vast opportunities for psychological manipulation. "These studies suggest that VR poses risks that are novel, that go beyond the risks of traditional psychological experiments in isolated environments, and that go beyond the risks of existing media technology for the general public," Madary and Metzinger say. They have developed recommendations for the use of VR, and call for long-term studies into the psychological effects of immersion. They also see a need for regulations regarding ownership and individuation of avatars, in addition to surveillance and data protection.
New Project to Watch Social Media for Signs of Mental Illness
CBC News (Canada) (03/03/16) Lauren O'Neil
University of Ottawa professor Diana Inkpen will receive $464,100 in funding from the Canadian government for a three-year project called "social Web mining and sentiment analysis for mental illness detection." Inkpen's team will explore the use of social media data in screening for individuals at risk of mental health issues. She has partnered with local data science technology company Advanced Symbolics to collect massive amounts of data from public sites such as Twitter, Facebook, and medical forums, and her team will use text-mining algorithms to find patterns in the data and predict what they mean. "Expressions of very negative emotions that are very strong, or appear a lot over longer periods of time, the algorithms can pick up," Inkpen says. "The algorithm learns from the data." The programs also can monitor how an individual's online activity changes over time. Inkpen says the goal is to create tools for flagging social media posts that can be used by doctors, psychologists, schools counselors, and research groups. If very angry or strange messages start appearing on social media, for example, the doctor of the patient writing them could receive an automatic alert. The team plans to roll out the tools in 2018.
Mobile Telephone Selects Best Route for Wheelchair Users
University of Twente (Netherlands) (03/03/16)
Researchers from the University of Twente's Center for Telematics and Information Technology (CTIT) Institute have developed technology that enables wheelchair users to map out safe and suitable routes and avoid roads and pathways that are unsuitable. The system processes and analyzes the mobile phone sensor data with which wheelchair users measure themselves. "Accelerometers and gyroscopes [a kind of compass] detect vibrations and changes in angle, and smartphones can register, process, and share these details with the sensor data analytics tool we developed," says University of Twente researcher Fatjon Seraj. The researchers are using the technology to determine the quality, suitability, and condition of roads, pathways, and buildings for wheelchair users, and are developing an app for the University of Twente campus. The system is based on the fact that wheelchairs are sensitive to vibrations, usually caused by uneven surfaces and differences in intensity and amplitude. These disturbances, which are transferred to the wheelchair's frame and then to the user, are picked up by the phone's sensors and analyzed by the data analytics tool.
Microsoft Research Scientist David Heckerman on How We Could Attack HIV Like Spam
The Washington Post (03/04/16) Meeri Kim
In an interview, Microsoft Research's David Heckerman describes applying machine learning to develop a vaccine to defeat human immunodeficiency virus (HIV) in the way a spam filter can foil spammers. As director of Microsoft Research's Genomics Group, Heckerman is using spam-filtering methods and scores of computers to analyze HIV's mutations to identify those that can kill the pathogen. Heckerman says the filter he invented in 1997 originally picked up on certain words, but when spammers easily thwarted it by varying the spelling of those words, his team rethought their strategy to scan the Internet and build a catalog of money-collecting sites to establish the likelihood of spamming if a message bore a link to one of those sites. "Spammers mutate their spam messages to work around our filters, and HIV mutates itself to avoid attack by our immune system," Heckerman notes. He says this observation prompted investigation to find vulnerable regions in the virus via machine learning. Heckerman notes a comparison of HIV carriers who were not severely ill with those who were showed viral sites marked by differences in where these attacks were occurring. A second machine-learning study simulating HIV's physical properties after mutating at various regions turned up mutations that destabilized the virus.
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.