Welcome to the May 5, 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).
|
Using AI-Enhanced Malware, Researchers Disrupt Algorithms Used in Antimalware
TechRepublic
Michael Kassner
May 4, 2017
Researchers at Peking University in China have found machine learning-based malware-detection algorithms cannot be used in real-world applications if they are easily bypassed by some adversarial techniques. The Chinese team reached this conclusion based on previous Google research demonstrating a technique to bypass malware-detection algorithms using altered information that maximized malware classification errors; this made it impossible for the detection algorithm to spot malware. The Peking University researchers built on the Google study by proposing the use of a generative neural network, called MalGAN, and altering the original samples to make input and output adversarial examples. The team trained a MalGAN generator to create adversarial examples that were capable of deceiving malware detectors. "Experimental results show that the generated adversarial examples are able to effectively bypass the malware detector," note Peking University researchers Weiwei Hu and Ying Tan.
|
Operating Smart Devices From the Space on and Above the Back of Your Hand
Saarland University
May 3, 2017
A team led by researchers at the Max Planck Institute for Informatics in Germany has developed a prototype input method called "WatchSense," which enables smart-device operation using the back of the hand and the space above it as a control "surface." The prototype employs a depth sensor, a miniaturized Kinect game controller, worn on the user's forearm about 20 centimeters from the watch. The sensor captures thumb and index finger movements on both the back of the hand and in the space over and above it in three dimensions. Software identifies the position and motion of the fingers so the user can run apps on devices. Max Planck's Srinath Sridhar says WatchSense was trained to distinguish between the fingers in real time via machine learning. The technology will be presented this month at the ACM Conference on Human Factors in Computing (CHI 2017) in Denver, CO.
|
Holography With the Wi-Fi Router
Technical University of Munich (Germany)
May 4, 2017
Researchers at the Technical University of Munich (TU Munich) in Germany have developed a holographic imaging process that depicts the radiation of a Wi-Fi transmitter to generate three-dimensional images of the surrounding environment. "Using this technology, we can generate a three-dimensional image of the space around the Wi-Fi transmitter, as if our eyes could see microwave radiation," says TU Munich researcher Friedemann Reinhard. Although processes that enable the localization of microwave radiation already exist, the novelty of the new system is that an entire space can be imaged using holographic processing of Wi-Fi or cellphone signals to a spatial resolution of a few centimeters. The researchers note future Wi-Fi frequencies will enable resolutions down to the millimeter range. They say further advancement of the technology could aid in the recovery of victims buried under an avalanche or a collapsed building by providing a spatial representation of destroyed structures.
|
Hand That Sees Offers New Hope to Amputees
Newcastle University (UK)
May 3, 2017
Researchers at Newcastle University in the U.K. are developing a new generation of prosthetic limbs that will enable the wearer to automatically reach for objects without thinking. The new bionic hand is fitted with a camera that instantaneously takes a picture of the object in front of it, assesses its shape and size, and triggers a series of movements in the hand. The researchers created the bionic hand using neural networks and showing the computer many object images. The researchers then taught the system to recognize the grip needed for each different object. The researchers programmed the hand to perform four different "grasps" by grouping objects according to the type of grip that would be needed to pick them up. The research is the first step toward the goal of developing a fully connected bionic hand that can sense pressure and temperature and transmit the information back to the brain.
|
Supercharging the Computers That Might Save the World
Umea University (Sweden)
May 4, 2017
Researcher Gonzalo Rodrigo at Umea University in Sweden has designed new methods and tools to manage high-performance computing (HPC) systems more efficiently so more advanced scientific research can be completed faster. "I have provided a better understanding of trends of current workloads and I have developed a general application-oriented scheduling model in HPC systems, a scheduling simulation framework to support future research on scheduling algorithms, and a scheduling technique for efficient execution of complex scientific workflows," Rodrigo says. He notes the results of this effort include two open source projects developed to enable future research on HPC scheduling. Rodrigo's work is motivated by the need to meet the growing demand for HPC systems with the capability of accommodating large volumes of data within research and enabling advanced simulations.
|
Computers Learn to Understand Humans Better by Modeling Them
Aalto University
May 3, 2017
Researchers at Aalto University in Finland, the University of Birmingham in the U.K., and the University of Oslo in Norway have developed a method that could help computers learn psychologically plausible models of individual people by observing them. The researchers showed that by observing how long a user takes to click on menu items, a program can infer a model that reproduces similar behavior and accurately estimates some characteristics of that user's visual system, such as fixation durations. The method is based on Approximate Bayesian Computation, which was developed to deduce very complex models from observations. The researchers say the method clears a path for automatic inference of complex models of human behavior from naturalistic observations. "The benefit of our approach is that much smaller amount of data is needed than for 'black box' methods," says Aalto's Antti Kangasraasio. The method could be useful in human-robotic interaction, or in automatically assessing individuals' capabilities.
|
Deep Learning Helps Scientists Keep Track of Cell's Inner Parts
University of Toronto
Jovana Drinjakovic
May 2, 2017
Researchers at the University of Toronto's Donnelly Center in Canada have developed DeepLoc, a deep-learning algorithm that can track proteins and help reveal what makes cells healthy and what goes wrong in diseased cells. The team says DeepLoc can recognize patterns in the cell made by proteins better and with much greater speed than previous computer vision-based approaches. In addition, the researchers say the algorithm can process images from other labs, illustrating its potential for wider use. Unlike computer vision that requires detailed instructions, DeepLoc learns directly from image pixel data, making it more accurate and faster. The researchers trained the algorithm on previously published data that shows an area in the cell occupied by more than 4,000 yeast proteins. The team notes DeepLoc was able to spot subtle differences between similar images, identifying 22 different classes of proteins, each representing distinct neighborhoods in the cell.
|
An Ostrich-Like Robot Pushes the Limits of Legged Locomotion
Technology Review
Will Knight
May 2, 2017
Researchers at the Institute for Human and Machine Cognition (IHMC) in Pensacola, FL, have developed the Planar Elliptical Runner, a two-legged robot that will help explore how mechanical design can be used to enable sophisticated legged locomotion. The robot's mechanical design provides dynamic stability as it runs, instead of relying on sensors and a computer to help balance itself. "We believe that the lessons learned from this robot can be applied to more practical running robots to make them more efficient and natural looking," says IHMC researcher Jerry Pratt. The Planar Elliptical Runner has a single motor that drives the legs; the elliptical motion of its legs together with its body shape provide inherent stability. The robot can run at 10 miles an hour, but IHMC researchers say if it were the size of a human, it would travel as fast as 30 miles per hour.
|
Artificial Intelligence Prevails at Predicting Supreme Court Decisions
Science
Matthew Hutson
May 2, 2017
A new study led by the Illinois Institute of Technology suggests artificial intelligence can outperform legal scholars in the prediction of U.S. Supreme Court rulings. The researchers built a general prediction algorithm based on the Supreme Court Database, drawing on 16 elements of each justice's vote, supplemented with other variables. For every year from 1816 to 2015, the team built a machine-learning "random forest" statistical model that reviewed all prior years and uncovered associations between case elements and decision outcomes. The model then examined the features of each case for that year to anticipate rulings, and was fed data about the rulings so it could update its approach and move on to the next year. The algorithm correctly forecast 70.2 percent of the high court's decisions and 71.9 percent of the justices' votes, while an earlier study found even knowledgeable legal scholars are only about 66-percent right in comparison.
|
"Valleytronics" Advancement Could Help Extend Moore's Law
UB News Center
Grove Potter
May 1, 2017
University at Buffalo (UB) researchers have discovered a new way to split the energy levels between the valleys in a two-dimensional semiconductor, which they say could lead to new, super-efficient computer chips. The researchers say the key to the breakthrough is the use of a ferromagnetic compound to pull the valleys apart and keep them at different energy levels. They note this leads to an increase in the separation of valley energies by a factor of 10 more than the one obtained by applying an external magnetic field. "Our new approach makes the valleys more accessible and easier to control, and this could allow valleys to be useful for future information storage and processing," says UB professor Hao Zeng. The research has pushed valleytronics a step closer to significantly extending Moore's Law, according to the researchers.
|
Smartphone Security Hole: 'Open Port' Backdoors Are Widespread
University of Michigan News
Nicole Casal Moore
May 1, 2017
"Open-port" backdoors in smartphone applications are far more susceptible to security hacks than previously assumed, according to a new study by researchers at the University of Michigan (U-M). The researchers found the vulnerability was most widespread in Android apps that enable users to share data across devices and link to phones from their computers. Using the OpenAnalyzer tool, the team scanned 24,000 popular mobile apps and identified 410 with serious flaws, and 956 different ways those flaws could be exploited. They also manually verified vulnerabilities in 57 apps, including popular file-transfer mobile apps with 10 million to 50 million downloads. More than half of the usage of open ports in the studied apps were found to be unshielded. U-M doctoral student Yunhan Hia says users should consider only using apps that share data from developers with solid reputations. The researchers also advise against using default passcodes for Android devices.
|
Supercomputers Assist in Search for New, Better Cancer Drugs
Texas Advanced Computing Center
Aaron Dubrow
May 1, 2017
Researchers increasingly are using supercomputers at the Texas Advanced Computing Center (TACC) to search for and evaluate new cancer drugs. One such project involves comparing plants used for chemotherapy and their genetic relations, with researchers using several TACC systems to develop a metabolic pathway database that functions as a community resource and can help identify new chemotherapeutic molecules. A second project focuses on applying artificial intelligence and high-performance computing-based virtual screening to accelerate cancer drug discovery and development. The team used TACC's Lonestar supercomputer to virtually screen and test drugs to determine which could bind and inhibit a cancer-causing enzyme. The largest atomic-level molecular modeling of a tumor-suppression protein to date comprised a third project using TACC resources. The effort helped specify new binding sites on the protein's surface where a small molecule could potentially be inserted to reactivate the inactive protein for more beneficial outcomes.
|
When Autonomous Machines Can Do More Than Just Vacuum
Scientific American
Rodney Brooks
May 1, 2017
Rapidly aging populations will drive growth in home-based robotics, according to Massachusetts Institute of Technology (MIT) Panasonic Professor of Robotics (emeritus) Rodney Brooks, who writes next-generation domestic robots are "going to have a much richer sensory world than existing home robots." Brooks says future robots also will benefit from more affordable and capable smartphone sensors, and the wide availability of low-power and miniaturized computing units. "Don't be surprised to see more silicon in phones over the next few years testing out [artificial intelligence] technologies such as deep learning," he says. Brooks also expects future domestic robots will feature additional sensors linked to their processors and in-device data handling. He cites MIT professor Dina Katabi's exploration of Wi-Fi radio signals as a possible tool for physiological readings of humans by domestic robots. "If our robots bypass the way that we experience the world with direct technological access to information, it will be hard for them to understand our limitations," Brooks notes.
|
|
