Welcome to the May 15, 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).
|
An Algorithm Summarizes Lengthy Text Surprisingly Well
Technology Review
Will Knight
May 12, 2017
Researchers at Salesforce have developed an algorithm that applies machine-learning techniques to accurately and coherently condense lengthy textual documents, technology which could impact fields such as law, medicine, and scientific research. The algorithm blends various strategies, including supervised learning, by being fed summary examples, while also applying an artificial attention mechanism to the text it is receiving and generating. The process ensures the system will not return too many repetitive strands of text, which has been an issue for other summarization programs. In addition, the system conducts experiments to produce its own summaries via reinforcement learning. Northwestern University professor Kristian Hammond lauds the Salesforce algorithm, but says it also illustrates the limits of solely relying on statistical machine learning. "We need a little bit of semantics and a little bit of syntactic knowledge in these systems in order for them to be fluid and fluent," Hammond says.
|
Physics May Bring Faster Solutions for Tough Computational Problems
University of Central Florida
May 12, 2017
Researchers at the University of Central Florida (UCF) and Boston University have developed a method to solve tough computational problems faster by applying statistical mechanics to create more efficient algorithms that run on traditional computers or a new type of quantum computer. UCL professor Eduardo Mucciolo and his collaborators have mapped computational problems onto a statistical model without phase transitions, defined on a two-dimensional lattice with each vertex corresponding to a reversible logic gate linked to four neighbors. "We configured it in such a way that every time these logic gates are satisfied, the energy is very low--therefore, when everything is satisfied, the overall energy of the system should be very low," Mucciolo says. The researchers believe the vertex model could solve sophisticated problems in machine learning, circuit optimization, and other computational challenges. The team also is investigating the model's application to the factoring of semi-primes.
|
You Really Should Know What the Andrew File System Is
Network World
Bob Brown
May 10, 2017
The creators of the Andrew File System (AFS) recently winning the 2016 ACM Software System Award demonstrates the value of knowing AFS as a foundational technology for many popular cloud computing techniques and applications. In an interview, AFS co-creator and Carnegie Mellon University professor Mahadev Satyanarayanan describes AFS as "the largest distributed file system that has ever been built and put to serious use." Satyanarayanan says the cloud-storage model and on-demand caching at the edge are AFS' most important contributions to modern cloud and enterprise computing environments. "All the data that is specific to a user is delivered on demand over the network," Satyanarayanan says. "Keeping in sync all the machines that you use becomes trivial. Users at organizations that deployed AFS found this an addictive capability." Satyanarayanan also notes AFS' architectural principles and deployment methods have significantly influenced many other systems developed over the past 10 years.
|
Social, Computer Scientists Want to Share Data on Group Behavior
NCSU News
Matt Shipman
May 9, 2017
Researchers at North Carolina State University (NCSU) have collaborated to develop a large dataset on how group behavior and technology influence decision-making. The work was made possible by a research partnership between NCSU and the U.S. National Security Agency. The data stems from an experiment conducted in which participants were asked to select a third-party presidential candidate for the 2016 election. Participants were given Internet access and asked to participate in group discussions to inform a draft report explaining their decision. Data being released included logs of study participants' Internet use, more than 85,000 screenshots illustrating the material viewed by participants, and a video of the screenshots for each group member who independently searched the Web for related information. The researchers say the work provides valuable insight on group behavior, and is useful for computer science questions related to predicting user behavior or machine learning. The research was presented last week at the ACM Conference on Human Factors in Computing Systems (CHI 2017) in Denver, CO.
|
New Georgia Tech Research May Help Combat Abusive Online Comments
Georgia Tech News Center
David Mitchell
May 9, 2017
Researchers at the Georgia Institute of Technology's (Georgia Tech) School of Interactive Computing have developed a technique to help Internet communities moderate abusive content. The Bag of Communities (BoC) technique uses large-scale, preexisting data from several online communities to train an algorithm to identify abusive behavior within a separate target community. The team identified nine different community types, five of which feature abusive behavior while four are helpful, positive, and supportive. Using linguistic characteristics from abusive and non-abusive communities, the researchers built an algorithm that can learn from comments and predict whether new posts are abusive within a target community. The team's dynamic model learns over time and can achieve 91.18-percent accuracy after being fed 100,000 human-moderated posts. Prior research into abuse detection and online content moderation had relied on data collected from within the same community, but in a BoC-based method, algorithms leverage out-of-domain data from other online communities.
|
Internet of Things Made Simple: One Sensor Package Does Work of Many
Carnegie Mellon News (PA)
Byron Spice
May 11, 2017
Researchers at Carnegie Mellon University's (CMU) Human-Computer Interaction Institute have developed a plug-in sensor package that tracks multiple phenomena in a room, collecting insights via machine-learning methods. "The idea is you can plug this in and immediately turn a room into a smart environment," says CMU's Gierad Laput. He notes the raw feeds from the package's nine sensors can be integrated and read by machine-learning algorithms in ways that can perceive numerous phenomena, including sounds, light, heat, temperature, and electromagnetic noise. Laput says the synthetic sensors can monitor the state of a device, while even more advanced sensors can deduce human activity. Plugging the units into a regular power socket makes batteries or special wiring unnecessary, although each room will likely require its own sensor platform. The researchers presented their "Synthetic Sensors" project last week at the ACM Conference on Human Factors in Computing Systems (CHI 2017) in Denver, CO.
|
FSU Technology Cracks, Fixes Passwords
Florida State University News
Kathleen Haughney
May 10, 2017
Researchers at Florida State University (FSU) have developed new software that evaluates password strength by trying to break it. The system takes a proposed password and generates guesses in the highest probability order. The more guesses it takes, the longer it takes an attacker to crack the password. The researchers say a strong password should be easy to remember and difficult to crack. "If our system can successfully crack a password, it will propose a password similar to the one submitted but with slight format variations, making it easier to remember," says FSU professor Sudhir Aggarwal. The researchers say the technology can provide support in a range of fields, especially for law enforcement attempting to crack encrypted files or hard drives. "Our program uses a more precise mathematical background compared to other applications, providing a more efficient process by generating password structures in highest probability order," Aggarwal notes.
|
Stanford Researchers Develop Crowdsourcing Software to Convene Rapid, On-Demand 'Flash Organizations'
Stanford News
Taylor Kubota
May 10, 2017
Researchers at Stanford University have developed software that integrates crowdsourcing's flexibility and the benefits of on-demand specialists to form "flash organizations." The researchers presented their work last week at the ACM Conference on Human Factors in Computing Systems (CHI 2017) in Denver, CO, detailing tests of their flash organization model. The team supported the model using the Foundry Web platform to help with generating the organization, hiring, task-tracking, and in-group communication. Foundry also featured a tool enabling members to request new roles or tasks as needed. To implement alterations, Foundry has an organizational chart that is used for all reconfigurations, including revising timelines of tasks, redefining roles, and hiring new workers. "By allowing anyone with an idea to go to an online marketplace, recruit all sorts of different experts on-demand, and bring their idea to life in a very short period of time, we're making innovation more feasible," says Stanford's Daniela Retelny.
|
Computer Accurately Identifies and Delineates Breast Cancers on Digital Tissue Slides
Case Western Reserve University
May 10, 2017
Researchers at Case Western Reserve University have developed a deep-learning computer network which they say is 100-percent accurate in determining whether invasive forms of breast cancer were present in whole biopsy slides. In addition, the researchers say the network correctly made the same determination in each individual pixel of the slide 97 percent of the time, rendering near-exact delineations of the tumors. The researchers trained the deep-learning network on 400 biopsy images downloaded from multiple hospitals, and afterward they presented the network with 200 new images from the Cancer Genome Atlas and University Hospitals Cleveland Medical Center. Network training took about two weeks, and identifying the presence and exact location of cancer took about 20 to 25 minutes each. "It will take time to get up to 20 years of practice and training of a pathologist to identify complex cases and mimics, such as adenosis," says Case Western professor Anant Madabushi.
|
Better Screenings Through Artificial Intelligence
Lehigh University
Lori Friedman
May 9, 2017
Researchers at Lehigh University say they have developed a cervical cancer screening technique that, based on an analysis of a very large dataset, could perform as well as, or better than, human interpretation or other traditional screening methods at a much reduced cost. The system, developed over 10 years by Lehigh professor Xiaolei Huang's team, is built on image-based classifier algorithms constructed from a large number of Cervigrams. The researchers hypothesized that algorithms could help enhance the precision in grading lesions by using visual information. Huang says she creates "techniques that enable computers to understand images the way humans do," and one of her primary interests is training computers to understand biomedical images. "Cervigrams have great potential as a screening tool in resource-poor regions where clinical tests such as Pap and [human papilloma virus] are too expensive to be made widely available," Huang notes.
|
Researchers Develop Transistors That Can Switch Between Two Stable Energy States
University of Illinois News Bureau
Lois E. Yoksoulian
May 9, 2017
Researchers at the University of Illinois at Urbana-Champaign (UIUC) have conducted a study in which they constructed optical and electrical bistable outputs from a single transistor, creating a feedback loop using a process known as electron tunneling that controls the transmission of light. The team says the new transistor could enable new devices and applications that have not been possible with traditional transistor technology. "Building a transistor with electrical and optical bistability into a computer chip will significantly increase processing speeds, because the devices can communicate without the interference that occurs when limited to electron-only transistors," says UIUC professor Milton Feng. The researchers have demonstrated electro-optical bistability at -50 degrees Celsius. Feng says the team also recently proved the device can operate at room temperature. "Any electronic device is virtually useless if it can't operate at room temperature," he notes.
|
Sound Over Silicon: Computing's Wave of the Future
Arizona Engineer
Jill Goetz
May 10, 2017
University of Arizona (UA) professor Pierre Deymier has conceived of a quantum computer analogue that processes information using sound waves instead of quantum particles. Deymier's team currently is building a phonon-based computer prototype, which Deymier says "has the power to change the world as we know it, not just for making more powerful computers, but for artificial intelligence, cryptography, and analysis of big data." Deymier believes phonons, incarnated as "phase-bits" or "phi-bits," can address a major problem in conventional quantum computers: the sensitivity of qubits to heat and similar environmental forces. The UA professor says lab experiments have demonstrated phi-bits can be generated at room temperature, and he also has shown multiple phi-bits can be assembled so they are inseparable. In addition, they can store information in a state of superposition. "I believe quantum computing with phononics will be feasible, possibly in the next 10 years," Deymier says.
|
|
