ACM TechNews
Association for Computing Machinery
Welcome to the November 30, 2015 edition of ACM TechNews, providing timely information for IT professionals three times a week.

Updated versions of the ACM TechNews mobile apps are available for Android phones and tablets (click here) and for iPhones (click here) and iPads (click here).


Teaching Machines How to Learn
ETH Zurich (11/30/15)

ETH Zurich and the Max Planck Society are collaborating on the creation of a new center that will study the theoretical principles of learning and apply them to robots and software. As robots and autonomous systems begin to operate in areas that will likely present them with new and novel challenges, it is important they have the ability, like humans, to learn how to cope with these new challenges. The new Max Planck ETH Center for Learning Systems was conceived as a place in which the next generation of scientists can be trained in the field of robotic learning and where collaborative research in the field can be carried out using a shared infrastructure. "We want to achieve a fundamental understanding of how people perceive, learn, and then react appropriately to situations," says Thomas Hofmann, a professor at ETH Zurich's Institute for Machine Learning and co-head of the new center. "If we can develop a better understanding of how learned aspects can be transferred between different tasks, we may be able to create artificial systems that learn in a similar way to living beings," says Bernhard Scholkopf, director of the Max Planck Institute for Intelligent Systems in Tubingen and co-head of the new center.

Microsoft Is Teaching Computers to See Like People
eWeek (11/28/15) Pedro Hernandez

Researchers from Microsoft and Carnegie Mellon University have combined computer vision, deep learning, and language understanding into a system that can analyze images and answer questions in the same manner as humans, according to Microsoft's Athima Chansanchai. The resulting model "applies multi-step reasoning to answer questions about pictures," Chansanchai says. The image-analysis system is based on earlier work by Microsoft on automatic photo-captioning technologies, which "helps train the computer to understand the image the way a person would," says Chansanchai. She notes the new system uses deep neural networks to absorb information as a human set of eyes and brain would, studying a scene's action and the relationships between multiple visual objects. Chansanchai says researchers from Microsoft's Deep Learning Technology Center are imbuing the system with attentional ability, and enabling it to concentrate on visual cues and infer answers progressively to solve problems. Microsoft hopes the technology will lead to systems capable of predicting human needs and providing real-time recommendations. The company also says the development of systems that answer questions based on visual input are essential to creating artificial intelligence tools.

Smile, Frown, Grimace, and Grin--Your Facial Expression Is the Next Frontier in Big Data
Smithsonian (12/15) Jerry Adler

Affectiva co-founder Rana el Kaliouby sees the use of computers to detect and interpret human facial expressions as the next logical step in the progression from keyboard to mouse to touchscreen to voice recognition. The field of "affective computing" seeks to close the communication gap between human beings and machines by adding a new mode of interaction, including the nonverbal language of smiles, smirks, and raised eyebrows, according to el Kaliouby. She notes emotions can guide or inform our rational thinking, but they are missing from the digital experience. "Your smartphone knows who you are and where you are, but it doesn't know how you feel," el Kaliouby says. She believes devices could control a car or things in the home such as lighting, temperature, and music more effectively if they know how humans feel. The core customers of Affectiva have been advertising, marketing, and media companies, but el Kaliouby believes the company's technology will be a boon to healthcare when it comes to getting feedback from patients on drug testing or treatment programs.

Why Ball Tracking Works for Tennis and Cricket but Not Soccer or Basketball
Technology Review (11/26/15)

Tracking balls in some sports--such as basketball, volleyball, and soccer--is significantly harder for machine-vision algorithms than it is for other sports. Swiss Federal Institute of Technology in Lausanne scientist Andrii Maksai and colleagues have outlined a new means for tracking balls that improves over other approaches. Such systems assume two dissimilar strategies: in one, ball movement is followed in three dimensions (3D) to predict likely future trajectories, which are narrowed down as more data becomes available. But this method tends to fail when the ball is hidden or when players engage with the ball in unforeseen ways. The second technique involves tracking the players and observing when they have the ball, with the motion of the ball assumed to follow the player and when it is transferred between players. However, this strategy can generate imprecise tracks when lacking physics-based limits on ball movement. "We explicitly model the interaction between the ball and the players as well as the physical constraints the ball obeys when far away from the players," says Maksai's research team. They have assessed the algorithm on video sequences of volleyball, soccer, and basketball games recorded on multiple cameras at different angles to produce a 3D model.

Computer Scientists Achieve Breakthrough in Pheromone-Based Swarm Communications in Robots
University of Lincoln (11/26/15) Elizabeth Allen

University of Lincoln researchers have developed a system that can replicate in robots all of the aspects of pheromone-based communication of insect swarms. Called COS-phi (Communication System via Pheromone), the system includes a low-cost open hardware micro robot and an open source localization system, which tracks the robots' trajectories and releases artificial pheromones. The researchers say the system is reliable and accurate. When using the system, the team's micro robots were able to follow the leader, or pheromone distributor, without any explicit direction or communication. "The system means that we can produce precise and high-resolution trails, control the diffusion, evaporation, and density of the pheromones, and encode individual pheromones using different colors," says Ph.D. researcher Farshad Arvin. The team has made the system available to robotics and artificial intelligence researchers. Research in swarm robotics has had applications in vehicle-collision sensors, surveillance technology, and video-game programming. "Nature is one of the best sources of inspiration for solutions to different problems in different domains, and this is why swarm robotics has developed into such an important area of study," Arvin says.

Computers Learn to Create Photos of Bedrooms and Faces on Demand
New Scientist (11/25/15) Jacob Aron

The capabilities of artificial neural networks are increasingly impressive, particularly when it comes to learning how to correctly identify objects. However, due to the opaque way such systems operate, researchers have a relatively weak understanding of how artificial neural networks do what they do. Recent efforts to investigate the processes artificial neural networks use to identify objects have yielded, among other things, Google's DeepDream project. Now researchers at Facebook and Boston-based machine learning firm indico are trying to determine how artificial neural networks work by asking them to "imagine" pictures. The focus of their effort is a type of artificial neural network called a generative adversarial network, in which one part of the system tries to create fake data to fool the other part that it is looking at training data. The idea is that by pitting the network against itself, it will learn to produce better images. The team asked the network to create images of bedrooms and faces and by tweaking their requests they were able to see the ways the network was developing concepts for different elements of a scene, such as a TV or a window, and how they relate to one another.

No Lens? No Problem for FlatCam
Rice University (11/23/15) David Ruth; Mike Williams

Rice University researchers Richard Baraniuk and Ashok Veeraraghavan have developed FlatCam, a thin sensor chip with a mask that replaces lenses in a traditional camera. FlatCam is equipped with algorithms that process what the sensor detects and converts the sensor measurements into images and videos. Veeraraghavan says FlatCams can be fabricated like microchips, with the precision, speed, and the associated reduction in costs. "Our design decouples the two parameters, providing the ability to utilize the enhanced light-collection abilities of large sensors with a really thin device," Veeraraghavan notes. The researchers say FlatCams could be applied to security or disaster-relief applications, as well as to flexible, foldable, wearable, and disposable cameras. The hand-built prototypes use off-the-shelf sensors and produce 512-by-512 images in seconds, but the researchers think the resolution will improve as more advanced manufacturing techniques and reconstruction algorithms are developed. "Smartphones already feature pretty powerful computers, so we can easily imagine computing at least a low-resolution preview in real time," says Carnegie Mellon University professor Aswin Sankaranarayanan.

Email Security Improving, but Far From Perfect
University of Illinois at Urbana-Champaign (11/18/15) August Schiess

Email security has improved significantly in the past two years, but widespread issues remain, according to a report from University of Illinois at Urbana-Champaign professor Michael Bailey in collaboration with colleagues at the University of Michigan and Google. The report notes networking protocols that underlie the Internet were not originally built to be secure, and security protocols were "bolted on" to the existing systems years later. Such measures are available to address security issues, but each individual server still has the choice whether to adopt the protocols, Bailey and colleagues found. The study also determined companies such as Google are now using these protocols, which have helped boost email security in recent years, but many other servers do not. The researchers measured the adoption of email security protocols at scale and also highlighted some of the implications of "bolted-on security." For example, the STARTTLS command is vulnerable to an attack that would force email exchanges to continue without encryption, the researchers note. "We found that there's a significant number of email exchanges in which there's an adversary between two mail servers who's trying to intentionally downgrade the communication," Bailey says.

Strategy Based on Human Reflexes May Keep Legged Robots, Prosthetic Legs From Tripping
Carnegie Mellon News (PA) (11/18/15) Byron Spice

Carnegie Mellon University (CMU) researchers have developed a robotic leg prosthesis that could help users recover their balance by using techniques based on the way human legs are controlled. CMU professor Hartmut Geyer says the control strategy was devised by studying human reflexes and other neuromuscular control systems, and it shows promise in simulation and laboratory testing, producing stable walking gaits over uneven terrain and better recovery from trips and shoves. The technology will be further developed and tested over the next three years thanks to a $900,000 U.S. National Science Foundation grant. "Our work is motivated by the idea that if we understand how humans control their limbs, we can use those principles to control robotic limbs," Geyer says. He thinks the research also could be applied to legged robots. The researchers evaluated the neuromuscular model by using computer simulations and a cable-driven device about half the size of a human leg. They found the neuromuscular control method can reproduce normal walking patterns and it effectively responds to disturbances as the leg begins to swing forward. "Robotic prosthetics is an emerging field that provides an opportunity to address these problems with new prosthetic designs and control strategies," Geyer says.

UCLA Computer Science Class Integrates Virtual World Into Reality
Daily Bruin (11/18/15) Nate Nickolai

Diana Ford, a lecturer in the University of California, Los Angeles (UCLA) computer science department, wants to develop gaming as a subfield of graphics at UCLA through her courses on virtual reality (VR) and artificial intelligence (AI). Her VR class currently is using gesture tracking to create its own VR games, and Ford's spring class presented games at the ACM SIGGRAPH conference in August. Ford says the interactions between players and AI gave rise to pre-patterns of coding, and developers around the world can use the pre-patterns to build games on an already developed base of code. Using Oculus Rift headsets, Ford says her spring class used Unreal Engine 4, a collection of game development tools, to generate three-dimensional worlds within which the students created original games with coding. The games focused on interactions between players and AI, which students produced using original programming methods in a variety of games. The pre-patterns produced by the spring class reduce coding time and enable game creators to focus on adding more aspects to VR games. Ford also presented the pre-patterns at the SIGGRAPH conference.

Seeking Data Wisdom
Berkeley Research (11/17/15) Wallace Ravven

Data wisdom is needed to make discoveries and assure the significance of the results of data-intensive research, says Bin Yu, a statistician and data scientist at the University of California, Berkeley. She describes the best of applied statistics as an essentially "soft lens" when working with large amounts of data, similar to a powerful telescope or precision gene microarray. Yu participated in Berkeley's "mind-reading" project in 2011, in which researchers used a type of magnetic resonance imaging (MRI) to detect indirect neuron firing at precise locations in the brain's visual processing area, and then determined the rough outlines of what experimental subjects were seeing in movie clips. Yu's team analyzed a torrent of functional MRI data to identify from thousands of movie clips the 100 frames that most likely matched a given voxel activity pattern, and then "averaged" these shapes to yield the outline of what subjects were seeing. Yu says only a powerful interlocking of science, computation, and statistics made this possible. "In computational neuroscience, it is important to gauge how much variation there is in the signals--in this case, how much of this variation is due to the movies and how much is due to 'noise,'" she says.

Stanford Students Put Computer Science Skills to Social Good
Stanford Report (11/19/15) Bethany Augliere

When Stanford University computer science (CS) undergraduate Lawrence Lin Murata heard a lecturer say last fall no campus organizations existed that used CS to make a positive social impact, he wondered if he could change that. A year later, Murata has created CS+-Social Good with the help of three other Stanford students and Keith Schwarz, their faculty sponsor. The group is focused on giving students opportunities to explore and practice their CS skills in the context of doing social good. It organizes speaking events and even organized a class this semester that focuses on bringing students together with nonprofit partners. Murata says the projects being worked on by the students in the Using Web Technologies to Change the World class "will reach over 25 million people by the end of the year." The projects include a group of students working with the government of Delhi, India, to create a website to track the progress of government programs, and a partnership with nonprofit group SIRUM to help connect institutions with surplus medication to safety-net clinics serving poor and uninsured populations. CS+-Social Good also plans to launch a new project this winter that will help four student teams identify and develop technological solutions for problems in areas such as healthcare and education.

Abstract News © Copyright 2015 INFORMATION, INC.
Powered by Information, Inc.

To submit feedback about ACM TechNews, contact:
Current ACM Members: Unsubscribe/Change your email subscription by logging in at myACM.
Non-Members: Unsubscribe