Welcome to the April 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
Who's the Michael Jordan of Computer Science? New Tool Ranks Researchers' Influence
Science (04/20/16) John Bohannon
Quantifying the influence a researcher or organization wields on subsequent scientific research is the purpose of an enhancement by the Allen Institute for Artificial Intelligence to its Semantic Scholar, an artificial intelligence (AI)-guided program originally designed to understand the content of published literature. Semantic Scholar ranks papers, authors, and institutions by a specific influence score. The highest-ranking computer scientist according to the program is AI researcher Michael I. Jordan of the University of California, Berkeley. The tool ranking scientists' influence was developed to enable the generation of an "influence graph" via machine learning. The technique was employed to train a statistical model that detects not only a scientific paper's distinct sections, but also the tone of how it is cited. The mode was improved by comparing its guesses with those of specialists who curated a sample of the papers. Allen Institute director Oren Etzioni says the system currently only measures "direct influence" between papers that cite each other, but future versions will account for the indirect influence of papers citing papers that subsequently cite other papers, and so on. Semantic Scholar currently is limited to papers on computer science, but Etzioni says the tool will expand to include neuroscience by the fall.
Computers Play a Crucial Role in Preserving the Earth
National Science Foundation (04/20/16) Aaron Dubrow
Cultivating a more sustainable world via computational methods is the goal of the field of computational sustainability founded by Cornell University professor Carla Gomes, with the help of an interdisciplinary team of programmers, theorists, applied mathematicians, economists, biologists, and environmental scientists. Gomes' team has developed basic computing tools for sustainability with a U.S. National Science Foundation (NSF) Expeditions in Computing award. One tool is a modeling method for wildlife corridors "that improved the movement of species over hand-designed corridors at a substantial fraction of the cost, considering trade-offs for multiple species," Gomes says. Another tool is a machine-learning system incorporating an active learning human-machine feedback loop, which helped Cornell researchers develop algorithms to compensate for biases in the distribution and misidentification of citizen scientist birders and accurately determine bird populations for conservation efforts. A further application of the team's research is designed to help nomadic African herders find forage by modeling their movements over a dryland ecology, and using a mobile phone app to crowdsource vegetation data combined with satellite imagery. The project's most significant achievement may be kickstarting the computational sustainability discipline and stimulating others to commit to sustainability initiatives. In January, NSF announced its support for the development of a Computational Sustainability Network comprised of various partners.
AOL-Cornell Tech Lab Pioneers New Content Technology
Cornell Chronicle (04/21/16) George Lowery; Melissa Osgood
The "Immersive Recommendations" technology announced by Cornell Tech and AOL on Wednesday is designed to address the challenge of how to engage users with relevant content when they first start using an online platform. The concept entails users opting into a tool that converts personal digital traces from one platform into content recommendations in another. "Users should be able to unlock their personal data for many purposes, including curating content based on their interests," says Cornell Tech professor Deborah Estrin. The tool, developed by Cornell Tech's Connected Experiences Lab, builds preference profiles of users according to their multichannel online activities. The researchers organized two demonstration systems, one that recommends articles and another that recommends meetups. The team found immersive recommendations enhanced recommendation performances by up to 57 percent, and 42 percent for news and meetups, respectively, over state-of-the-art methods. "Capturing signals across as many channels as possible leads to better targeting and syndication of content, especially in mobile settings," says AOL's William Pence. "We are interested in how these techniques can be applied to create more relevant and 'immersive' ad experiences, specifically in video and in [virtual reality]." In its current deployment, Immersive Recommendations lets users control which services they permit to share and use their data.
A Simple Way to Hasten the Arrival of Self-Driving Cars
Technology Review (04/20/16) Will Knight
Researchers at Oxford University's Mobile Robotics Group have compiled a massive public dataset for self-driving vehicles culled from thousands of hours of data from a single 10-kilometer (slightly more than six miles) stretch of road over a 12-month period. The project emphasizes how the sharing of such data could expedite the deployment of driverless cars. The researchers monitored the sort of variation self-driving cars will need to contend with every day, such as moving vehicles, cars parked in different ways, and lighting variations. Oxford's Will Maddern says longer-term changes are tracked as well, including "construction, roadworks, seasonal changes in vegetation, etc." Maddern notes one of the purposes for the dataset was to gain insights into at what points autonomous vehicle systems would malfunction, especially those that depend on precise mapping. Oxford's data on showing variation along a single route will be more extensive than that collected by Google and Tesla. Massachusetts Institute of Technology professor John Leonard envisions large-scale and long-duration datasets providing "a huge boost to the rate of progress." He believes the sharing of such data by self-driving vehicle developers could accelerate the technology's practical use.
MIT News (04/21/16) Larry Hardesty
Massachusetts Institute of Technology (MIT) researchers have developed a decentralized planning algorithm for teams of robots that factors in stationary obstacles as well as moving obstacles. The researchers say the algorithm requires significantly less communication bandwidth than existing decentralized algorithms, but preserves strong mathematical guarantees that the robots will avoid collisions. They tested the algorithm in simulations involving squadrons of minihelicopters and the algorithm came up with the same flight plans a centralized version did. The idea behind the algorithm is that each robot, on the basis of its own observations, maps out an obstacle-free region in its immediate environment and passes that map only to its nearest neighbors. When a robot receives a map from a neighbor, it calculates the intersection of that map with its own and passes that along; the technique reduces both the size of the robots' communications and their number, because each robot communicates only with its neighbors. However, each robot ends up with a map that reflects all of the obstacles detected by all of the team members. The maps also consider time, which is how the algorithm accounts for moving obstacles. The four-dimensional map describes how a three-dimensional map would have to change to accommodate the obstacle's change of location over a span of a few seconds.
ETH Researchers Print Wild Robotic Beings
ETH Zurich (04/20/16) Peter Ruegg
Researchers from ETH Zurich, Disney Research Zurich, and Carnegie Mellon University are developing a software tool that enables users to quickly custom-design a robot on a computer. Users create a basic skeleton for their robot and design its movement, using simple motion goals such as "walk forward" or "turn left." The software provides immediate feedback on the resulting motion, as predicted by simulation. Once the user is satisfied with the design, the tool generates three-dimensional (3D) building plans for the robot, including the connecting parts that house the electric motors. The program includes standard sizes of various commercially available motors, and users select the one that matches in order to obtain the connecting parts. Consumers would use a 3D printer to fabricate the robot and then assemble it by hand. The researchers used the software to create a simple five-legged robotic insect that could move forward and sideways, but the robot could not identify objects and could not be controlled remotely, functions that were outside the scope of the project.
Physicists Build 'Electronic Synapses' for Neural Networks
MIPT News (Russia) (04/20/16)
Researchers from the Moscow Institute of Physics and Technology (MIPT) have developed prototype "electronic synapses" based on ultra-thin films of hafnium oxide (HfO2) for potential use in computing systems that function in a similar manner to biological neural networks. The team combined "analog" memristors in matrices to form unique nanostructures, and HfO2-based devices exhibit soft or reversible electrical breakdown under an applied external electric field. Such devices frequently use only two different states encoding 0 and 1 logic, but to simulate biological synapses required a continuous spectrum of conductivities. "In the metal-ultrathin oxide-metal structure, charged point defects, such as vacancies of oxygen atoms, are formed and move around in the oxide layer when exposed to an electric field," notes MIPT's Sergey Zakharchenko. "It is these defects that are responsible for the reversible change in the conductivity of the oxide layer." The analog memristors were employed to simulate learning mechanisms of biological synapses, including long-term potentiation or depression of a link between two neurons. The researchers also demonstrated spike-timing-dependent plasticity by applying an electric signal to replicate the signals in biological neurons. Such outcomes verified these elements could be a prototype of an electronic synapse, which could serve as the platform for artificial neural network hardware.
How MIT is Using Ripple to Push Blockchain Research Beyond Theory
CoinDesk (04/19/16) Michael del Castillo
The Massachusetts Institute of Technology (MIT) has entered into an alliance with distributed ledger tech startup Ripple to bring its blockchain research out of theory and into practice. MIT Connection Science's David Shrier says this implementation should draw a broad spectrum of researchers, with the ultimate goal of the effort being enterprise data projects such as blockchain and financial services. MIT currently is running a validator, a server that confirms transactions on the network on which the XRP digital asset resides, for the Ripple Consensus Ledger, its permissioned distributed ledger system. Ripple's solid position in the finance sector partly influenced MIT's decision to partner with the company, according to Shrier. He notes MIT now has about 36 blockchain-related projects in development, and that number should grow. The project can be traced to MIT's long-term interest in open source initiatives developed by its Internet Trust Consortium. In June 2015, the consortium released the results of its "Enigma" research focusing on how developers could build a decentralized cloud platform using blockchain. The effort uses an outside blockchain to regulate who can access data and the identities of those users.
Russia, U.S. Get Closer to Universal Memory
EE Times (04/18/16) R. Colin Johnson
A joint effort by researchers at the Moscow Institute of Physics and Technology (MIPT), the University of Nebraska, and Switzerland's University of Lausanne seeks a "universal" non-volatile memory composed of a ultra-thin ferroelectric film grown on silicon. "We grow polycrystalline [rather than epitaxial] alloyed hafnium-zirconium [Hf-Zr] oxide films, which retain their ferroelectric properties down to thicknesses of under three nanometers," notes MIPT researcher Andrei Zenkevich. The compatibility of this material with silicon substrates enables existing complementary metal-oxide semiconductor fabrication plants to transition to the Hf-Zr oxide material. The next step for the researchers is to construct prototype ferroelectric tunnel junctions with the films to show they can be used in actual memory chips. "The work to demonstrate the so-called tunneling electroresistance effect in a prototypic memory device is underway now," Zenkevich says. "Judging from pulsed measurements of the polarization reversal, the prospective write time is within the nanosecond range." Ferroelectric tunnel junctions are extremely small, yet can retain their values indefinitely without consuming power. The material could potentially be adapted to function like an artificial neuromorphic synapse so it can be incorporated into both conventional computers and future cognitive computers.
Root Is a Little Robot on a Mission to Teach Kids to Code
Wired (04/18/16) Davey Alba
Smartphone Users Are Redefining Privacy in Public Spaces
American Friends of Tel Aviv University (04/18/16)
Tel Aviv University (TAU) researchers say "dynamic visibility," in which technological surveillance is combined with personal information volunteered by individuals online, diminishes privacy. TAU's Tali Hatuka says any time a person uses location-aware devices, driving or dating apps, or checks in on Facebook, they are eroding their own privacy. The TAU researchers found some differences among sharing preferences in different types of digital spaces, but these paled beside the willingness of participants to share their locations with their social networks. The researchers developed Smart-Spaces, an Android application that combines smartphone-based surveys with the online tracking of locations and phone app usage. The app was installed for 20 days on the phones of TAU students, who answered context-based surveys in the course of their daily routines. Hatuka notes not only did more than 73 percent of participants share their locations as they answered the surveys, but "there was a correlation between the kind of space they were in...and their willingness to provide information, with a higher willingness to share location and other information when the subject was in public spaces."
The Professor Who Helps People Find Things
The Irish Times (04/21/2016) Dick Ahlstrom
Dublin City University professor Alan Smeaton has won a Royal Irish Academy Gold Medal for his work with big data as founding director of the Insight Center for Data Analytics. "I build information systems that help people to find what they are looking for," Smeaton says. Among the topics the center focuses on are the enhancement of memory retrieval in people with dementia, computer learning, and personal sensing via more efficient use of data sources, with an emphasis on deriving value from information databases. In terms of memory retrieval, Smeaton says the work entails accumulating a person's memories, activities, and everyday things using wearable cameras to construct a digital footprint. He notes people with short-term memory difficulties can access this archive to re-invoke a context. Smeaton also is studying training a computer to describe video and picture content, which could augment automatic captioning. He notes systems currently can recognize a traffic cone with about 60-percent accuracy, a road with 75-percent accuracy, and a truck with 80-percent accuracy. "If you combine things you get an ensemble effect that gives you interdependencies, and interdependencies can make good descriptors for digital content," Smeaton says.
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.