Welcome to the October 30, 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."
|
|
|
What Virtual Reality Can Teach a Driverless Car
The New York Times
Cade Metz
October 29, 2017
Major companies are testing autonomous car software within virtual reality (VR) simulations of cities, mainly to spot and correct operational flaws without imperiling actual passengers. Scientists also are devising techniques to enable cars to learn new behavior from VR, accumulating skills faster than human engineers can program them. Toyota Research Institute CEO Gill Pratt says Toyota is using images of simulated roadways to train neural networks, and they are sufficiently similar to the real world to reliably educate the systems that operate the cars. Other firms are exploring reinforcement learning to gain skills within virtual environments via trial and error. One issue concerns the fact that because these programs learn by analyzing more information than anyone ever could, it can be difficult to audit their behavior and understand why they arrive at specific decisions. Nevertheless, machine learning is expected to be vital to the continued progress of autonomous vehicles in the coming years.
*May Require Free Registration
|
From Small to Not So Pixel-Perfect Large
Max Planck Institute for Intelligent Systems
October 27, 2017
Researchers at the Max Planck Institute for Intelligence Systems in Germany used artificial intelligence (AI) technology to develop a high-definition version of a low-resolution image. The researchers note the technology to create a large-sized image from a low-resolution image is known as single-image super-resolution (SISR), which has been studied for decades with limited results. The Max Planck researchers proposed a new approach, EnhanceNet-PAT, to give images a realistic texture when magnified from small to large using machine learning. When AI is applied, an adaptive algorithm for upsampling the images learns from experience to improve the result. The researchers say the new technology is more efficient than other SISR systems because EnhanceNet-PAT does not attempt pixel-perfect reconstruction; instead, the new AI system aims for faithful texture synthesis. By detecting and generating patterns in a low-resolution image and applying these patterns in the upsampling process, the team says the system adds extra pixels to the low-resolution image accordingly.
|
Best-Ever Algorithm Found for Huge Streams of Data
Quanta Magazine
Kevin Hartnett
October 24, 2017
A multi-institutional research team last fall created a near-perfect streaming algorithm that operates by recalling only enough of what it has seen to relate what it has observed most frequently. The researchers note a key principle of previous streaming algorithms is dividing them into sub-algorithms. "Instead of spending 50 million units of time looping over the entire universe, you only have four algorithms spending 100 units of time," says Harvard University's Jelani Nelson. However, he notes the main drawback to this approach is the difficulty of extracting the right small numbers to recombine in order to yield the correct big number. Nelson and colleagues instead package each two-digit block with a small tag that consumes little memory while permitting the algorithm to properly reassemble the two-digit pieces. The researchers also use an expander graph to boost the likelihood of correctly connecting the chain of two-digit blocks into the right number.
|
Artificial Intelligence to Evaluate Brain Maturity of Preterm Infants
University of Helsinki
Paivi Lehtinen
October 20, 2017
Researchers at the University of Helsinki in Finland have developed a method that uses artificial intelligence to assess a preterm infant's brain maturity from an electroencephalogram (EEG). A large volume of EEG data on preterm infants was fed into a computer, and the machine-learning software calculated hundreds of computational features from each measurement without requiring intervention from a physician. A support vector machine algorithm helped combine these features into a reliable estimate of the EEG maturational age of the infant. At the end of the study, the researchers compared the EEG maturational age with the known true age of the infant and found the true age and computer-generated calculation were within two weeks of one another in more than 80 percent of the cases. "This method gives us a first-time opportunity to track the most crucial development of a preterm infant, the functional maturation of the brain, both during and after intensive care," says Helsinki professor Sampsa Vanhatalo.
|
Singapore Teaches Its Seniors to Code
CNet
Aloysius Low
October 30, 2017
Libraries in Singapore are hosting special versions of the global Hour of Code movement, which is helping seniors aged 50 and older learn how to program in the Swift language alongside student volunteer instructors. The event is part of a larger initiative driven by the Singapore government to make the country's senior citizens more familiar with technology. The global Hour of Code project is designed to help students of all ages learn coding and computer science, and it claims to have already reached out to 100 million students in 180 countries. In Singapore, the participating seniors spend about an hour learning the basics of programming, first getting an avatar moving around before moving on to loop integrations. Although the Hour of Code's participants are unlikely to develop advanced coding skills, the one-hour training sessions could help spark interest in a different hobby to keep their minds active and alert.
|
Computer Learns How to Imagine the Future
Albuquerque Journal
Garrett Kenyon
October 27, 2017
Researchers at the Los Alamos National Laboratory are modeling biological neural networks on supercomputers so machines can learn about their surroundings, interpret data, and make human-like predictions. The lab's Trinity supercomputer enables a different approach to large-scale neuromimetic computing, with the latest milestone being a "sparse prediction machine." This construct features two neural networks executed in parallel: a predictive network and a network that emulates representations of future video frames it cannot view directly. The machine was exposed to thousands of eight-frame video sequences, each showing a particular object, in much the same way a child accumulates visual experience. Each neuron in the network learned to represent a specific visual pattern, and the sparse prediction machine then used the representations learned by the individual neurons while concurrently developing the ability to predict the eighth frame from the preceding seven. Los Alamos plans to advance the use of sparse prediction machines in neuromorphic computing.
|
There Are Three Kinds of Consciousness, and Computers Have Mastered One, Says Study
ScienceAlert
Mike McRae
October 27, 2017
A study by neuroscientists offers insights into whether computers may become conscious by deconstructing consciousness into three categories. The researchers describe the first category (C0) as encompassing problem solving by the brain without awareness, and computers already can achieve this level, as manifested in the advent of driverless vehicles. The researchers say the second category (C1) "refers to the relationship between a cognitive system and a specific object of thought." On this level, that object is chosen for global processing, transferring it from a narrow relationship into one that can be manipulated under various contexts. The third category (C2) covers "meta-cognition," or a sense of knowing what we know, and the researchers note although C1 can occur without C2, and vice versa, neither system as yet has a machine-intelligence equivalent. They suggest C2 technology could be developed by applying probability to decision-making, or by retaining a meta-memory to establish a connection between what is and is not known.
|
This Group of U of T Computer Science Researchers Are Decoding Ciphers With AI
U of T News
Nina Haikara
October 26, 2017
Researchers at the University of Toronto in Canada are using a neural network to decrypt text. The FOR.ai team's project uses plain text, or English, and cipher text as two distinct languages. The researchers say the network both reads and establishes connections between the two without any additional translation support. "None of [the algorithm] is hard-coded or relying on a human's knowledge of language," says FOR.ai member Aidan Gomez. "We came up with an architecture than can infer those mappings independently." Gomez notes the algorithm can crack the Vigenere cipher, in which a hidden key is known only to the sender and recipient. "It's part of the goal of getting closer and closer to the complexity of unsupervised language translation itself," he says, calling the project a demonstration of "a neural network's capacity to build up a really strong model of language, and then apply that to drawing connections between two abstract languages."
|
Funding Supports New UChicago-Based Computer Science Education Initiative
UChicago News (IL)
Louise Lerner
October 26, 2017
University of Chicago professor Diana Franklin has received a $2.5-million U.S. National Science Foundation grant in support of the Computing for ANyONe (CANON) lab. CANON is a new computer science education initiative divided into three projects that aim to teach more children the basics of computer science earlier in their school careers. "The digital divide is real and growing, and we want to provide educators with easy, inclusive, and tested tools to integrate these concepts into their classrooms," Franklin says. The CANON lab wants to create content that does not rely on teachers to have previous experience with coding and programming. CANON projects include the Comprehending Code program for reading and understanding code, while the Learning Trajectories for Everyday Computing project addresses computational and mathematical thinking for third- to fifth-graders studying fractions. Finally, the Scratch Encore program is a collaboration between Chicago Public Schools and the University of Chicago STEM Education center.
|
Women and Robots, 'There's Just No Trust'
Western Sydney University
Farah Abdurahman
October 24, 2017
Researchers at Western Sydney University in Australia conducted experiments to determine whether a trust dynamic could be established between humans and robots, and found women were much less likely than men to trust humanoid machines. Western Sydney roboticist Chris Stanton compared how three levels of robot gaze--averted, constant, and situational--affected the probability of participants accepting the robot's advice when they disagreed on the correct answer. "Women became noticeably uncomfortable and more guarded when the robot stared at them, but men behaved very differently, with frequent eye gaze from the robot appearing to have no effect," Stanton notes. "In the experiment where the staring robot disagreed with the participant's response, women stuck with their gut instincts and did not change their answer despite coercion by the robot." Stanton says the research is valuable in the push for human-robot collaboration, noting, "Social robots must be capable of fostering the trust and confidence of people they interact with."
|
UTSA Receives $5 Million to Support New Cybersecurity Education Pipeline
USTA Today
Joanna Carver
October 24, 2017
Researchers at the University of Texas at San Antonio (UTSA) have received a $5-million grant from the U.S. National Science Foundation to develop the Center for Security and Privacy Enhanced Cloud Computing (C-SPECC), a multidisciplinary center fostering education and research in cybersecurity and cloud computing. The researchers say the center will be a pipeline to create well-trained professionals in the industry and strengthen San Antonio as a cybersecurity hub. "The center will also give UTSA students an unparalleled learning opportunity to conduct research alongside the university's nationally recognized experts in cyber, cloud, computing, and analytics," notes UTSA president Taylor Eighmy. C-SPECC will include the UTSA colleges of science, engineering, education, and business, which will work together to help the facility become nationally recognized for excellence in research and innovation in secure cloud computing. In addition, the center plans to boost participation among underrepresented minorities in high-tech computing and to pursue innovative research-based educational strategies.
|
NSF Awards NCSA Funds for a Deep-Learning Research Instrument
HPCwire
October 24, 2017
Researchers at the University of Illinois at Urbana-Champaign's (UIUC) National Center for Supercomputing Applications (NCSA) have been awarded more than $2.7 million from the U.S. National Science Foundation to develop an instrument to expand research into deep learning. "This deep-learning instrument will bolster current relevant deep-learning research communities here at the University of Illinois, allowing researchers to leverage deep learning more than they ever could before," says NCSA director Bill Gropp. He says the grant will open the way for new industry-academia collaborations, and the framework of the system architecture and the instrument will be made public, in order that other researchers can build off of NCSA's breakthroughs. The grant also will support university researchers as they investigate new systems and algorithms for machine learning and data analytics that extract actionable knowledge from massive datasets, says UIUC professor Roy H. Campbell.
|
|
