Welcome to the December 13, 2017 edition of ACM TechNews, providing timely information for IT professionals three times a week.
ACM TechNews mobile apps are available for Android phones and tablets (click here) and for iPhones (click here) and iPads (click here).
To view "Headlines At A Glance," hit the link labeled "Click here to view this online" found at the top of the page in the html version.
The online version now has a button at the top labeled "Show Headlines."
|
|
|
This Machine-Learning Algorithm Can Turn Any Line Drawing Into ASCII Art
Motherboard
Daniel Oberhaus
December 12, 2017
Osamu Akiyama at Osaka University in Japan says he has developed a machine-learning neural network that can render any kind of line drawing in ASCII that is very similar to human artists' ASCII artwork. Akiyama trained the network using 500 ASCII drawings derived from the popular Japanese message boards 5channel and Shitaraba. He says the challenge stems from the fact that much handmade online ASCII art does not cite the original image the ASCII work is based on, and so the algorithm is unable to learn how a line drawing is translated into text. Akiyama addressed this with a network designed by other scientists to clean up rough sketches so ASCII art can be reverse-engineered to the original drawing. Once the network produced estimates of the original line drawing, they were used as input to train the network to learn which characters were used to create the ASCII image.
|
Stellenbosch Graduate's Braille-to-Text Breakthrough
TimesLive (South Africa)
Petru Saal
December 11, 2017
Researchers at Stellenbosch University in South Africa have developed software that can instantly convert Afrikaans text to braille, a breakthrough they say could have a significant impact on teaching. The program, developed by Stellenbosch's Pallier Gerber, was written in such a way that there are different levels of complexity between Grade 1 and Grade 2 braille. "A teacher can now prepare text and then decide‚ with the click of the mouse‚ on which level they want to present it‚" Gerber says. The program could be used to help with lesson preparation and significantly reduce the time it takes to convert reading material from Afrikaans to braille. The program also enables teachers to compare what a student wrote in braille with the Afrikaans equivalent on a computer screen and immediately give feedback. "This means I will now be able to teach like any other teacher in a mainstream school‚" notes Stellenbosch vice principal Hannes Byleveldt.
|
AI Experts Caution Senate Against Heavy Regulation
CIO Dive
Alex Hickey
December 12, 2017
At a hearing yesterday of the U.S. Senate Committee on Commerce, Science, and Transportation, industry and academic experts warned against overly regulating artificial intelligence (AI) technology, and called for open data policies. Senate leaders questioned the gathered experts on the problem of bias in AI and machine learning, with the experts testifying that bias is multilayered, as it can be introduced at many levels. In addition, the dearth of diversity in the AI workforce compounds the issue, according to Princeton University professor Edward Felten. The experts also agreed education and understanding play an essential role in the creation of a well-rounded AI policy and culture, starting with better science, technology, engineering, and math training at early ages, as well as in higher education. The Information Technology and Innovation Foundation's Daniel Castro differentiated between AI policies designed to expedite adoption--which are positive--and regulation, which he described as likely misguided.
|
Computer Scientists Develop a Simple Tool to Tell If Websites Suffered a Data Breach
University of California, San Diego
Daniel Kane
December 12, 2017
Researchers at the University of California, San Diego (UCSD) say they can detect when websites are hacked via a new tool, Tripwire, which monitors the activity of email accounts associated with those sites. Tripwire has a bot that registers and creates accounts, each with its own unique email address, on thousands of sites; each email account and the site account associated with that address are assigned the same password. The researchers found almost 1 percent of the sites they tested suffered a data breach during the 18-month study, which means that tens of millions of websites could be breached yearly, according to UCSD Ph.D. student Joe DeBlasio. In addition, the researchers found popular sites were just as likely to be hacked as unpopular ones, meaning out of the top 1,000 most visited sites on the Internet, 10 are likely to be hacked every year. The researchers presented Tripwire last month at the ACM Internet Measurement Conference (IMC 2017) in London.
|
New Silicon Structure Opens the Gate to Quantum Computers
Princeton University
Catherine Zandonella
December 11, 2017
Researchers at Princeton University have built a silicon gate capable of controlling quantum behavior between two electrons, or quantum bits (qubits), with very high precision. The researchers say the new gate paves the way for making sophisticated, multi-qubit devices using technology that is less costly and easier to manufacture than other approaches. "We knew we needed to get this experiment to work if silicon-based technology was going to have a future in terms of scaling up and building a quantum computer," says Princeton professor Jason Petta. By constructing the silicon quantum devices in Princeton's Quantum Device Nanofabrication Laboratory, the researchers could keep the spins coherent for relatively long intervals. The team layered aluminum wires onto a highly ordered silicon crystal, and by temporarily lowering the energy barrier, they enabled the electrons to share quantum information via entanglement. The researchers showed they could use the first qubit to control the second qubit, indicating the structure served as a controlled NOT gate.
|
Diagnosing Performance Problems in Supercomputers
Government Computer News
Stephanie Kanowitz
December 12, 2017
Researchers at Sandia National Laboratories and Boston University (BU) have spent more than a year developing the Lightweight Distributed Metric Service (LDMS), a framework to automatically monitor and diagnose performance issues in supercomputers. Using LDMS to diagnose supercomputer problems should help systems administrators allocate resources and schedule jobs to maximize performance. The team says they used supervised machine learning, writing programs to reproduce known anomalies that would likely affect a Cray XC30m supercomputer at Sandia and BU's Mass Open Cloud system. With LDMS, the supercomputer compiled more than 700 metrics each second for each computer, and the cloud collected about 50 metrics at two- or three-second granularity. Sandia's Vitus Leung notes the difference stems from the "noisiness of the data on the BU cloud, because it's not nearly as dedicated." The researchers collated statistical characteristics of the data, filtering it to about 10 percent of the raw data, which was fed to machine-learning algorithms.
|
Security and Privacy for Democracy Panel
CCC Blog
Helen Wright
December 11, 2017
Panelists at the Computing Community Consortium (CCC) Symposium in October discussed how security and privacy must be supported to ensure democratic principles are protected. Roger Dingledine with the Tor Project stressed the importance of transparency in the online Tor community, upheld by its free and open source nature. Meanwhile, the U.S. Census Bureau's Simson L. Garfinkel noted the bureau is building a public disclosure avoidance system for the 2020 Census. Garfinkel said the Bureau will publish the source code, and he noted the system will rely on infusing formally private noise. Rice University's Dan Wallach said the security/privacy risk of electronic voting can be addressed with voting machines using cryptographic and other verification methods, including paper records. Meanwhile, the University of Massachusetts, Amherst's Phillipa Gill cited the challenge of measuring and tracking censorship because of the Internet's decentralized design, and noted scientists are developing different forms of censorship across diverse content.
|
Twitter Can Reveal Our Shared Mood
University of Bristol News
December 11, 2017
Researchers at the University of Bristol in the U.K. have analyzed mood indicators in the text from 800 million anonymous messages posted on Twitter, and found the tweets reflect strong patterns of positive and negative moods over a 24-hour day. The researchers examined the use of words relating to positive and negative emotions on Twitter over the course of four years. Although previous studies showed a circadian variation for positive and negative emotions, the new Bristol study also was able to differentiate specific aspects of anger, sadness, and fatigue. The researchers found distinct patterns of positive emotions and sadness between the weekends and weekdays, as well as evidence of variation of these patterns across the seasons. "The patterns that our research revealed for the positive emotions and sadness showed more variability in response to these changing conditions, and higher levels of interaction with the onset of sunlight exposure," says the University of Bristol's Fabon Dzogang.
|
Revolutionizing Electronics Using Kirigami
ScienceDaily
December 8, 2017
Researchers at the Toyohashi University of Technology in Japan have developed an ultrastretchable bioprobe using Kirigami designs, which enables the device to follow the contours of spherical and large deformable biological samples, such as heart and brain tissues. "The remarkable feature of Kirigami is that rigid and unstretchable materials can be rendered more stretchable compared to other elastomer-based stretchable materials," says Toyohashi's Yusuke Morikawa. In addition, Morikawa notes the bioprobe's strain-stress property is very low compared to that of elastomer-based stretchable devices because the stretching mechanism is based on an out-of-plane bending of the thin film. "Our preliminary studies on Kirigami-based parylene films by microelectromechanical systems technology exhibited high stretchability of 1,100 percent," says Toyohashi professor Takeshi Kawano. The team thinks the bioprobes also can be used to explore tissues and organs that exhibit time-dependent alterations in their surface and volume due to growth or disease.
|
Neural Networks Are Learning What to Remember and What to Forget
Technology Review
December 8, 2017
Researchers at the University of Leuven in Belgium and Facebook AI Research have demonstrated that the strategy biological systems employ to learn and to forget also can be applied to artificial neural networks. Leuven's Rahaf Aljundi based his team's learning method on the concept of repeated synchronized firing of neurons to strengthen connections and make them harder to overwrite, by quantifying the outputs from a neural network and tracking their sensitivity to changes in their connections. The team says this establishes the most important and preservation-worthy network parameters, giving the network "memory aware synapses." "We show that a local version of our method is a direct application of Hebb's rule in identifying the important connection between neurons," the researchers note. They say that by improving their version of Hebbian learning, machines should be enabled to be more flexible in their learning, and thus capable of better adaptation to the real world.
*May Require Free Registration
|
AI Researchers Are Trying to Combat How AI Can Be Used to Lie and Deceive
Quartz
Dave Gershgorn
December 8, 2017
Artificial intelligence (AI) researchers gathered at last week's Neural Information Processing Systems (NIPS 2017) conference in Long Beach, CA, to discuss measures against AI's use for deceit and disinformation. One workshop concentrated on tactics in which adversarial examples are used to fool AI into seeing something that does not really exist. Workshop co-organizer Tim Hwang says the potential for such abuse of AI is growing, "especially if you think the inputs to do machine learning are getting lower and lower over time." Hwang is concerned about AI-powered disinformation making it virtually impossible for large populations to distinguish reality from fiction, or whether trust of online content will ultimately only be possible via technological authentication. NIPS workshop co-organizer Bryce Goodman warns of "systems that are trained to exhibit features of human intelligence but are fundamentally different in terms of how they process information. We're trying to show what hacks are possible and make it public."
|
Seeing Through Walls of Unknown Materials
Duke Pratt School of Engineering
Ken Kingery
December 6, 2017
Researchers at Duke University have developed a technique that exploits a wall's symmetry to see through it using a narrow band of microwave frequencies without any advance knowledge of what material the wall is made from. Since walls are generally flat and uniform in all directions, they distort the microwaves in a symmetrical fashion. "We wrote an algorithm that separates the data into parts--one that shows circular symmetry and another that doesn't," says Duke postdoctoral researcher Okan Yurduseven. The new method identifies the data that does not have any symmetry, using only a single frequency to scan because it reduces the number of interference patterns created by the wall. During testing, the researchers analyzed the data and removed the symmetrical patterns, and could make out objects placed behind the walls. "We envision combining this technique with a machine-vision system that someone could move over a wall to see what's inside," says Duke professor Daniel Marks.
|
Neurons Have the Right Shape for Deep Learning
Canadian Institute for Advanced Research
Juanita Bawagan
December 5, 2017
Researchers at the Canadian Institute for Advanced Research (CIFAR), the University of Toronto in Canada, and Google DeepMind say they have developed an algorithm that simulates how deep learning could work in human brains. It is based on neurons in the neocortex, which governs higher orders of thought; the network shows certain mammalian neurons have the shape and electrical properties that are well-suited for deep learning, and this breakthrough represents a more biologically realistic way of how real brains could perform deep learning. The researchers used their knowledge of the neurons' structure to build a model that similarly received signals in segregated compartments, enabling simulated neurons in different layers to collaborate and achieve deep learning. IFAR's Blake Richards says this field of research could see an environment in which neuroscience discoveries help to develop new artificial intelligence (AI), and AI can help interpret and understand experimental data in neuroscience.
|
|
