Welcome to the May 10, 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).

Image of lung with cancerous masses Million-Dollar Prize Hints at How Machine Learning May Someday Spot Cancer
Technology Review
Will Knight
May 9, 2017


Researchers in China developed a contest-winning algorithm that could reduce the number of errors in computed tomography (CT)-based lung cancer diagnoses. The contest required the teams to use only 2,000 CT images to train their machine-learning algorithms, and the winners used a neural network and rigorously annotated images to supply more data points. In addition, the winning team employed another dataset and split the challenge into a two-part process by first identifying nodules and then diagnosing cancer. "We think that explicitly dividing this problem into two stages is critical, which seems also to be what human experts would do," says Tsinghua University's Zhe Li. The contest, sponsored by Booz Allen Hamilton and offering a $1-million prize, was designed to help overcome existing software's lack of reliability in detecting lung cancer by applying deep-learning technology, says the U.S. National Cancer Institute's Keyvan Farahani.

Full Article
Harnessing the Potential of Big Data to Improve the Security of Internet of Things Devices
A*STAR Research
May 9, 2017


Researchers at the Agency for Science, Technology and Research (A*STAR) in Singapore have proposed algorithms that generate and store a very small piece of secret information within an Internet of Things (IoT) device. The information can concisely represent the whole history dataset, and can be retrieved for authentication. The researchers note this two-step system can be compatible with IoT devices with low computation and small memory. The researchers found using the data exchanged between the device and the server has very interesting leakage-resilience properties. Since data is constantly generated by the IoT device and sent to the server, the history dataset is always growing. As a result, hackers would have to steal a significant amount of data over an extended period of time, becoming more open to detection in the process.

Full Article
147 Teams Announced in $5M IBM Watson AI XPRIZE
The Huffington Post
Craig Zamary
May 3, 2017


One hundred forty-seven teams, representing 22 countries, are advancing in the $5-million IBM Watson AI XPRIZE, a four-year global competition to develop and demonstrate how humans can collaborate with artificial intelligence (AI) technologies to address some of the world's greatest challenges. This is XPRIZE's first "open" competition, in which teams defined their own goals and will create AI applications that solve some of humanity's most daunting challenges across a range of fields. Some of the teams are focusing on solutions that may challenge underlying assumptions about AI, including encoding human ethics within AI, imbuing AI with human social norms, and utilizing AI to understand human emotional cues. Other teams will focus on domain-specific solutions, including health and wellness, learning and human potential, civil society, space and new frontiers, shelter and infrastructure, energy and resources, and the planet and the environment.

Full Article

Photo of green lasers Laser-Controlled Touch Displays Near Reality
EE Times Asia
May 9, 2017


Researchers at the University of Nagoya in Japan have discovered a phenomenon called the photodielectric effect, which they think could be applied to the development of laser-controlled touch displays. "A photo-capacitor provides a novel way for operating electronic devices with light," says Nagoya's Hiroki Taniguchi. "It will push the evolution of electronics to next-generation photo-electronics." Taniguchi and colleagues made their discovery by experimenting with an intrinsic photodielectric effect in a ceramic by shining a light-emitting diode (LED) onto the ceramic and quantifying its dielectric permittivity, which rose even at high frequencies. However, unlike prior experiments that used the extrinsic photodielectric effect, the material exhibited solid insulation properties. The researchers say this means the LED is directly changing the dielectric permittivity of the material, and not elevating conductance. Taniguchi says it remains uncertain how the intrinsic photodielectric effect operates, but he thinks it may be related to defects in the material.

Full Article
NASA Challenges Coders to Speed Up Its Supercomputer
The New York Times
Jacey Fortin
May 8, 2017


The U.S. National Aeronautics and Space Administration (NASA) has issued a challenge to programmers to boost the performance of its Pleiades supercomputer in the execution of computational fluid dynamics, offering thousands of dollars in cash prizes. NASA's Michael Hetle says it currently takes days or weeks for Pleiades to yield an output. The two-fold challenge involves the crowdsourcing of big-picture, strategic concepts to re-envision and improve the FUN3D code, and conceive of tactical revisions to accelerate Pleiades' efficiency. FUN3D uses the Fortran programming language, which Franklin W. Olin College of Engineering professor Allen Downey says is uniquely suited to function with algebraic algorithms that apply to physical motion. NASA says the competition is designed to help it analyze FUND3D's performance bottlenecks and identify possible alterations that might trim overall computing time. Downey notes the contest offers NASA a cost-effective way to tackle a major problem.

Full Article

Photo of hands typing using a braille keyboard A Touchable Tablet to Guide the Visually Impaired
Swiss Federal Institute of Technology in Lausanne
Laure-Anne Pessina
May 5, 2017


Researchers at the Swiss Federal Institute of Technology in Lausanne, Switzerland (EPFL) have developed a tablet to help visually impaired people navigate unfamiliar places. The device quickly forms shapes and relief maps that users can explore with their fingers using their sense of touch. In addition, the tablet could be used to help visually impaired schoolchildren learn subjects such as geometry or mathematics. The tablet consists of 192 tiny buttons that can move up and down in milliseconds, almost instantaneously creating patterns for users to feel and interpret. Each button contains a tiny magnet placed between two coils and two thin layers of steel. The EPFL team says any of the buttons can be moved up or down by generating a local magnetic field. The technology was presented this week at the ACM Conference on Human Factors in Computing Systems (CHI 2017) in Denver, CO.

Full Article
CMU Creating Touchpads With Can of Spray Paint
CMU News
Byron Spice
May 8, 2017


Researchers at Carnegie Mellon University (CMU) have developed "Electrick," technology that can turn a wide variety of surfaces into touchpads. Electrick works like a can of spray paint, adding conductive coatings or materials to objects or surfaces, or it can be used to create objects using conductive materials. The researchers showed that by attaching a series of electrodes to the conductive materials, they can employ electric field tomography to sense the position of a finger touch. "For the first time, we've been able to take a can of spray paint and put a touchscreen on almost anything," says CMU professor Chris Harrison. Electrick depends on the "shunting effect," and the researchers were able to localize where and when such shunting transpires by affixing electrodes to the edge of an object or conductive coating. Electrick will be presented this week at the ACM Conference on Human Factors in Computing Systems (CHI 2017) in Denver, CO.

Full Article
Engineering Research Focuses on Bringing Efficiency to Network Processes
UNM Newsroom
Kim Delker
May 3, 2017


Researchers at the University of New Mexico (UNM) have used complex numerical modeling to instill efficiency in networks. The team tackled the challenge of reducing energy expenditure when trying to control a large distributed system. UNM's Isaac S. Klickstein says of particular significance was the finding that the effort can be streamlined by focusing the goal of the control action toward only the most important network elements instead of tracking all of them. "We ended up seeing that we get essentially the same type of behavior by removing control action goals as previous papers got by increasing the number of control action locations," Klickstein notes. UNM professor Francesco Sorrentino says the research could enable the control of systems that may have been impossible using previous methods. Klickstein says the relevance of the team's research should grow as technological advances lead to ever-greater interconnection of systems.

Full Article
Inquisitive Bot Asks Questions to Test Your Understanding
New Scientist
Matthew Reynolds
May 8, 2017


Researchers at Cornell University say they have developed an artificial intelligence (AI) system that has learned to ask questions of its own. The machine-learning algorithm is capable of reading a passage of text and coming up with questions that check a person's understanding of the topic. The researchers trained a neural network on more than 400 Wikipedia articles and 100,000 questions about those articles sourced from crowdworkers. The software learned to recognize patterns that linked questions back to their source text. The researchers then presented the AI with portions of Wikipedia articles it had not yet seen and found it was able to produce a question for every sentence it reads. The researchers say the next step is to train the system to ask questions only about sentences that contain statements. "Not all sentences are question-worthy," says Cornell Ph.D. student Xinya Du.

Full Article

A robot holding a rifle The Military Is Using Human Brain Waves to Teach Robots How to Shoot
Defense One
Patrick Tucker
May 5, 2017


The U.S. Army is partly funding research by DCS Corp. and the Army Research Lab to teach robots to shoot targets with greater precision by feeding datasets of human brain waves to a neural network. "You have to figure out...in that dynamic environment we have in the military world, how do we retrain this [deep]-learning process from a systems perspective," says Army chief scientist Thomas Russell. The researchers want the neural network to facilitate trials in which a computer can easily understand when a soldier is assessing targets in a virtual situation, instead of investing much time educating the system on how to organize different individuals' data, eye movements, P300 responses, and other factors. The researchers say their goal is a neural network capable of instantaneous, continuous, and real-time learning by observing soldiers' brain waves and eye movements.

Full Article
Conference Connects, Empowers Women in Computing
Penn State News
Jennifer Cifelli
May 3, 2017


The recent Grad Cohort Workshop sponsored by the Computing Research Association for Women (CRA-W) hosted nine graduate students from Pennsylvania State University's College of Information Sciences and Technology (IST) to help them network and gain knowledge from peers and experts. Workshop attendees were offered advice from industry and academic professionals about navigating their graduate education and insights on subjects ranging from networking and mentorship to research and dissertations. The IST students agreed that cultivating strong mentoring relationships and peer networks were among the most persuasive reasons for attending, as well as being helpful to them as women in a traditionally male-dominated discipline. "When I meet so many women in the same shoes as me, I feel we are closely related; I feel more confident pursuing a career in computing," says IST student Haining Zhu. Meanwhile, IST student Xinye Zheng notes the event "offers us a platform to communicate and share common experiences."

Full Article
Cornell CIS and Adobe Collaboration Creates AI Photo Tool
Cornell Chronicle
Leslie Morris
May 4, 2017


Researchers at Cornell University and Adobe have developed Deep Photo Style Transfer, software that can transpose the look of one photo onto another using neural networks to ensure the details of the original image are preserved. "What motivated us is the idea that style could be imprinted on a photograph but it is still intrinsically the same photo," says Cornell professor Kavita Bala. The researchers say the major breakthrough involved preserving boundaries and edges while still transferring the style. They used deep machine learning to add a neural network layer that pays close attention to edges within the image, such as the border between a tree and a lake. "The method we came up with is surprisingly very effective," Bala says. "It has definitely captured people's imaginations as a way to stylize photos in a more dramatic way."

Full Article
Computers That Can Smell
The Scientist
Kerry Grens
May 1, 2017


Eighteen teams participated in an IBM Research Dialogue for Reverse Engineering Assessments and Methods (DREAM) challenge to develop algorithms that can estimate how humans will perceive a particular odor from its molecular properties. Rockefeller University's Andreas Keller compiled a dataset of odor perceptions, specifying more than 4,800 molecular characteristics for each compound. The challenge's goal was designing the best model from data on 69 smells to anticipate their scent profiles. A linear-model algorithm placed second by taking different parts of each molecule and predicting how each bit would smell. IBM Research's Pablo Meyer suggests this may reflect something basic concerning olfaction and how odors interact with receptors. Meyer says instead of an entire molecule matching a distinct receptor, perhaps it engages with numerous receptors, each of which responds to these molecular subunits. "A nice step would be to see how that relates to the binding of odor molecules to odor receptors," he notes.

Full Article
Computer Graphics & Interactive Techniques Conference
 
ACM Digital Library
 

Association for Computing Machinery

2 Penn Plaza, Suite 701
New York, NY 10121-0701
1-800-342-6626
(U.S./Canada)



ACM Media Sales

If you are interested in advertising in ACM TechNews or other ACM publications, please contact ACM Media Sales or (212) 626-0686, or visit ACM Media for more information.

To submit feedback about ACM TechNews, contact: technews@hq.acm.org