Faculty Profile: Fang Finds Fulfillment in AI for Social Good

Kids today grow up with computers, and some of them even take programming classes before they hit middle school. But when Fei Fang was in seventh grade, that kind of thing was unheard of. That is, until her school hired a new teacher with a background in computer programming — a teacher who offered an extra class for students who wanted to learn Pascal.It's lucky for Carnegie Mellon University, and arguably the whole artificial intelligence field, that Fei Fang took that class. Because it started her down a path that would eventually lead to where she is today: an assistant professor in the School of Computer Science's Institute for Software Research who is taking the world by storm with her AI-for-social-good agenda.Let's back up a bit, though.Fang grew up about an hour outside Shanghai, and her experience learning Pascal in middle and high school inspired her to enroll in Tsinghua University's electrical engineering program, where she quickly learned that she preferred software applications to her coursework in hardware. With that in mind, she started applying to Ph.D. programs focused on image processing, but nothing seemed super exciting to her. Then something happened."During that time, I learned that a student from Tsinghua had gone to a Ph.D. group that was working on game theory and AI for infrastructure security, and that their work was being deployed to protect the Los Angeles International Airport," Fang said. "I thought that was interesting and cool, so I contacted the person who would become my Ph.D. advisor, told him about my background and how interested I was in his work. Then I got the chance to work with him."To take advantage of that opportunity, Fang left China for the University of Southern California, where she joined the group led by Milind Tambe, the Helen N. and Emmett H. Jones Professor in Engineering, and now the director of the Center for Artificial Intelligence in Society. There, her work revolved around the idea of using game theory for security and sustainability.Fang says that the earliest discussions related to game theory were simply people trying to determine the best strategies for playing games. For example, if you're playing "Rock, Paper, Scissors," the best strategy is to play randomly. Game theory grew out of that, and into a field that models strategic behaviors — in all sorts of domains — as a game."As long as the situation includes multiple, self-interested agents and they are strategically interacting with each other, we can use game theory to model it," Fang said.One of the problems she studied as part of her Ph.D. research — and that she continues to investigate at CMU — is how to protect wildlife from poachers. Many wildlife reserves tasked with protecting animals like elephants, tigers and rhinos have limited resources and rely on a small number of enforcement officers to patrol vast reserves, putting them at a huge disadvantage when it comes to catching poachers. Fang hopes using game theory combined with machine learning techniques will help change all that."As you can imagine, if law enforcement agencies patrol in a deterministic way — say they always take a certain path on Monday morning — this can be easily exploited by the attacker, the poacher," Fang said. "We have to help them randomize."Through the Protection Assistant for Wildlife Security (PAWS) project, Fang and her team built models of both the law enforcement and the poachers' behavior and applied game theory techniques to help determine the optimal patrol routes for law enforcement. But these models don't go far enough to help the agents do their jobs. According to Fang, game theory models generally assume that people are perfectly rational and choose the action that leads to the highest utility. But we all know people aren't perfectly rational. So Fang and her team needed to actually learn about their human "players" from the data they collected.That's when machine learning enters the picture.Fang's team used the data their collaborators from conservation agencies such as the World Wildlife Federation and the Wildlife Conservation Society collected on both patrol and poaching activity to build behavioral models that help predict what will happen in the future. Then they applied game theory techniques to that data to create the best patrolling strategy. As more data enters the model — as patrollers follow the randomized routes and provide feedback to the researchers on poaching evidence — the model will begin to learn the poacher's behaviors and develop the best randomized patrol routes to thwart them.Stopping poachers in their tracks is just one way Fang has used machine learning and game theory — and, thus, artificial intelligence — for social good. Her work has also helped randomize the U.S. Coast Guard's routes for protecting the Staten Island Ferry from attack, detect illegal mining sites from satellite images and protect fisheries."AI for social good has become hot, probably in the last two years," Fang said. "My hope is that as we bring this theme up, we can inspire more people to work on more social-good challenges and guide AI toward a direction that can benefit society instead of leading to more and more concerns for humans."One way she's fueling that inspiration is through a course, aptly titled "Artificial Intelligence Methods for Social Good." Designed for both undergraduate and graduate students, the course provides students with a firm foundation in AI techniques like machine learning, game theory and mechanism design, sequential decision-making, and planning and optimization. But it doesn't stop there. It also introduces students to challenges facing the fields of healthcare, social welfare, security and privacy, and environmental sustainability, and how they can apply AI techniques to solve those challenges. The course also contains an ethics component, and features guest lecturers from other departments and colleges."We see many people who know artificial intelligence techniques, but don't want to handle the real challenges we face in the world today. They just want to work on a well-established problem and come up with new algorithms," Fang said. "I hope that with this course, we can equip students with AI methods that have broad applications to social-good challenges."Fang said that CMU is the natural choice for her efforts, because of its reputation for developing real tools to solve real problems."I do want to make an impact with my work. I want to reach out. I want to deploy our research, like we've done before," she said. "Being at CMU provides me with the resources to collaborate with different agencies and see our work used in the real world. It allows me to make that impact."

Researchers Reinvent the Wheel for Vehicles of the Future

Shape-Shifting Tires, Digital Driving Assistants Could Enable Safe Driving Over All Kinds of Terrain

Wheels that transform into tracks on the fly and a digital assistant that helps drivers find the safest, surest route across steep terrain — or even does the driving at times — are technologies that could change expectations of what vehicles can do. Both this reinvention of the wheel and creation of a hybrid human/computer driver were accomplished at Carnegie Mellon University's National Robotics Engineering Center as part of a Defense Advanced Research Projects Agency (DARPA) program called Ground X-Vehicle Technologies, or GXV-T. Shape-Changing Wheel Technology Could Do "Amazing Things" The reconfigurable wheel-track can function as either a wheel or a triangular track to enable vehicles to operate at high speed on roads, or to traverse diverse off-road terrains. It can transform from one mode to the other in less than two seconds while the vehicle is in motion. "This shape-changing locomotion technology could enable vehicles to tackle a wide array of terrains at surprising speed," said Dimi Apostolopoulos, a CMU Robotics Institute senior systems scientist who led the project at NREC. "Based on the testing we've done so far, we would expect such a vehicle to do amazing things." DARPA's GVX-T program aims to reduce the need for armor by making combat vehicles faster, more maneuverable and capable of operating in a wide variety of environments. Apostolopoulos said the shape-shifting wheel-track has a number of potential civilian applications as well, including uses in forestry, mining and construction, and heavy equipment. In testing to date, vehicles equipped with the reconfigurable wheel-track have achieved 50 miles an hour in wheel mode and almost 30 mph in track mode. The device has transformed from wheel mode to track mode at speeds as high as 25 mph and from track mode to wheel mode at speeds of around 12 mph. The core of the wheel-track concept is to adjust the size of the contact patch — the area of the wheel-track in contact with the ground — based on surface type and wheel loading. By reducing the contact patch on smooth surfaces, it is possible to achieve higher speeds. Increasing the contact patch maximizes traction, much like a snowshoe, and enables the vehicle to maneuver safely on soft soils. Today's Humvees do something similar by increasing tire pressure on smooth, dry surfaces and reducing tire pressure in off-road situations, Apostolopoulos said. Other groups have done something similar to NREC's reconfigurable wheel-track, though those previous efforts usually have required halting the vehicle to transform from one mode to the other, Apostolopoulos noted. The ability to make these transformations on the fly, he added, is a critical requirement for vehicles that must handle changing terrain at high speed. The reconfigurable wheel-track has a rubberized tread that sits atop a frame that can change shape. An early version used electric motors to change the shape of the frame, but the researchers soon realized they could transform the shape passively, using the speed of the wheel-track itself to power the shape change, rather than the complicated system of motors. In its latest manifestation, the spinning wheel is transformed into a track by extending a Y-shaped support, which pushes the frame into a triangular shape. Simultaneously, application of a brake to stop the wheel from spinning causes the transmission to automatically shift from turning the wheel to turning a set of gears that drives the track.

Grant Helps Carnegie Mellon, University at Buffalo Improve Transit Access

Renewed Federal Grant Supports Research on Transportation for People With Disabilities

A 10-year partnership between Carnegie Mellon University and the University at Buffalo (UB) to advance physical access and public transportation for people with disabilities has been extended for another five years. The two universities' joint Rehabilitation Engineering Research Center (RERC) on Accessible Public Transportation has received a five-year, $4.6 million grant from the U.S. National Institute on Disability, Independent Living and Rehabilitation Research. The center develops ways to empower consumers, manufacturers and service providers in the design and evaluation of accessible transportation equipment, information services and physical environments. "This new cycle of funding will include projects on making autonomous vehicles accessible and leveraging commercial artificial intelligence agents, such as Alexa and Google Home, to support people with disabilities," said Aaron Steinfeld, the center's principal investigator and an associate research professor at Carnegie Mellon's Robotics Institute, where he works on human-robot interaction and intelligent transportation systems. Steinfeld co-directs the center with Jordana Maisel, the director of research activities at UB's Center for Inclusive Design and Environmental Access (IDeA). The IDeA Center improves the design of environments and products by making them more usable, safe and appealing to people with a wide range of abilities. In the past, RERC researchers at the Robotics Institute have used a publicly deployed mobile app testbed, Tiramisu Transit, to examine how to best gather and provide information for a rider's trip. In parallel, UB has evaluated physical bus characteristics and the built environment to inform new vehicles, construction and rulemaking. "The team has also had an active collaboration with industry, which we plan to continue and expand over the next five years," Maisel said. In parallel with the RERC is the Disability Rehabilitation Research Project (DRRP) on Robotics and Automation for Inclusive Transportation, a five-year effort awarded to Carnegie Mellon in the fall of 2017 in coordination with the U.S. Department of Transportation Accessible Transportation Technologies Research Initiative (ATTRI). This project, which includes the University of Washington, is conducting research and development on seamless transportation assistance from cloud-based autonomy and shared robots located in and around transportation hubs. For more information on the RERC on Accessible Public Transportation, visit the organization's website.

Advancing Robotics Will Boost U.S. Manufacturing Competitiveness

A new report by the Boston Consulting Group (BCG), co-authored by Robotics Professor Howie Choset, argues that a thriving domestic robotics industry will help to enhance America's competitiveness in the global economy, especially now that tensions are rising over global trade. The report suggests that robotics would have the greatest impact in five sectors — aerospace, apparel, electronics, machine shops and motor vehicles — that collectively represent 29 percent of U.S. manufacturing imports, 10 percent of labor costs and 36 percent of the nation's trade imbalance. Robotics would help U.S. companies cope with a shortage of labor and declining productivity growth. Greater use of robots would improve competitiveness with China, which is significantly subsidizing robotics as part of its Made in China 2025 initiative, the authors said. Tasks that cry out for robotics solutions include the identification, tracking and movement of parts and tools in factories and warehouses; technology to support frontline supervisors; assembly and installation of wiring and wire harnesses; precision handling and assembly of small components; and translating schematics into production without the need for a human intermediary. To achieve this, technical progress is essential in perception, sensing and situational analysis, enabling robots to accommodate a variety of lighting conditions and work with a variety of materials so they can be reconfigured rapidly. Robots will need to be more dexterous and, in contrast to today's industrial robots, safe to work alongside people. The report emphasized that the workforce will require training in new skills and will need to continually learn. Standardization of the metrics used for robotics and of robotics hardware and software will help manufacturing executives better evaluate technical solutions and keep manufacturers from getting locked into their vendors. Choset authored the report along with BCG's Justin Rose, Vlad Lukic, Claudio Knizek, Tom Milon and Alex Melecki. Support for the study also was provided by the Advanced Robotics for Manufacturing Institute, which was co-founded by Choset along with Gary Fedder, CMU vice provost for research, and George Darakos, director of partnerships for the School of Computer Science.

Pfenning Wins NIDA Avenir Award

Grant Will Fund Research on Genetic Factors Associated With Substance Abuse

The National Institute of Drug Abuse (NIDA) has awarded Andreas Pfenning, assistant professor in the Computational Biology Department, an Avenir Award for genetics or epigenetics research, which the agency presents to early stage investigators who propose highly innovative studies. The award includes a five-year, $1.5 million grant that will sponsor the Pfenning laboratory's efforts to discover genetic factors associated with the predisposition to addiction. "The development of new treatments for substance use disorders has remained difficult due to the complexity of the neural circuits and the underlying genetic mechanisms," Pfenning said. Research suggests that regulatory regions of the human genome, particularly regions known as enhancers that increase the likelihood that particular genes are transcribed, are associated with complex brain disorders, such as addiction. "Despite the importance of enhancer regions in the brain, the computational and experimental tools to study their function are still in their infancy," Pfenning added. His lab will develop a map that links genetic variation associated with substance abuse to neural enhancer function and establish a computational and experimental framework to study any combination of enhancers and genetic variations in different brain regions, cell types and animal models. The Avenir Awards — "avenir" is the French word for "future" — represent NIDA's commitment to supporting researchers who represent the future of addiction science. Pfenning was one of six recipients in 2018. Pfenning, who joined the faculty in 2016, received a 2018 Sloan Research Fellowship and last year was named part of an all-star international research team by the Cure Alzheimer's Fund. He earned his bachelor's degree in computer science at Carnegie Mellon, and his Ph.D. in computational biology and bioinformatics at Duke University.

SCS Sophomores Share Their CMU Firsts: Semesters, Years and Experiences

When we last spoke to School of Computer Science students Trevor Arashiro, Chris Choi, Lauren Morgenthaler, and Peter Wu, they were first-years, entirely new to the undergraduate computer science program, the city of Pittsburgh, and most importantly, college. Last fall, these talented students provided insightful snapshots of their first semesters in the School of Computer Science. Now — with a year of college under their belts — they're back with more.Trevor Arashiro"The most memorable SCS experience I had my freshman year was building my 15-112 term project. The project simulated the Wean elevators, finding the most efficient algorithm for picking people up. It exposed me to just how complicated elevators were, despite originally thinking that I could write a simulated one in less than 200 lines of code."Back in high school, my CS classes didn't teach me much, but the courses here provided me with everything I could handle and then some. I learned that there were some classes I wasn't at the top or middle of, but instead was at the dead bottom. Freshman year truly taught me what it means to be an SCS student, and the responsibility it carries."Something I'd tell my freshman-year self is what I would tell every incoming freshman: don't start slacking on your work. A lot of things are different from high school, and there's a lot less external motivation pushing you to do your work. Self-discipline is a must, and deadlines are DEADLINES. Which means even 10 seconds after the deadline, the assignment is still late!"

Carnegie Mellon Qatar Team Wins in Oman Programming Competition

Two CMU-Q Teams Advance to the Arab Regionals

A team of three Carnegie Mellon University in Qatar (CMU-Q) students won the Oman Collegiate Programming Competition on Oct. 17 at Sultan Qaboos University in Muscat. Mohammed Yusuf Ansari, Mohammed Nurul Hoque and Akhyar Kamili were the only team to solve nine of 12 problems. A second CMU-Q team, comprising Sameer Ahmad, Ishaq Yusuf Haj Hasan and Hari Krishna, placed fifth in the competition. A total of 52 teams from 22 institutions participated. Both CMU-Q teams have now qualified to compete at the Arab Collegiate Programming Competition regional competition. Michael Trick, dean of CMU-Q, congratulated the students for the win. "We are very proud of how these six students represented CMU-Q, and the state of Qatar, at this competition in Oman. Our students work very hard and produce high-quality work, and it is gratifying to share their talent with the international community." "I would also like to congratulate the entire team of CMU-Q computer science professors, and in particular Giselle Reis, who advised the students, for their dedication to computer science education," he continued. "This win reflects the quality of their teaching and their commitment to student learning." All six students are part of CMU-Q's Computer Science Program, which follows the School of Computer Science's curriculum.

Blum, Reddy Will Present Keynote Lectures at Microsoft Asia Event

Distinguished Career Professor of Computer Science Lenore Blum, and University Professor of Computer Science and Robotics Raj Reddy will be keynote speakers at the Computing in the 21st Century Conference, Nov. 6–7, in Beijing. The conference, celebrating its 20th anniversary this year, is Microsoft Research Asia's largest annual event. Blum's address will focus on her work with husband Manuel and son Avrim on developing a computer architecture that could enable conscious artificial intelligence — machines that truly experience feelings, rather than simulate them. Reddy will review the tools, techniques and advances of AI over the past half-century, separating hype from reality, and will discuss two types of intelligent agents: one that will help people work faster and more efficiently, and another that helps people do tasks previously impossible for humans. Harry Shum, a Robotics Institute alumnus who is executive vice president of Microsoft's AI and Research Group, and Peter Lee, former head of the Computer Science Department and now corporate vice president of Microsoft Research, also will be keynote speakers.

Sherry Named Rising Star in Computer Networking, Communications

Justine Sherry, assistant professor in the Computer Science Department, is among 10 researchers chosen as this year's N²Women Rising Stars in Computer Networking and Communications. N²Women – or Networking Networking Women — is a community of researchers that promotes diversity and fosters connections among underrepresented women in the computing subfield of networking and communications. Its annual list of rising stars recognizes women who have had an impact on the field early in their careers. Sherry's research interests include middleboxes, networked systems, measurement, cloud computing and congestion control. Her recent research focuses on new opportunities and challenges arising from the deployment of middleboxes — such as firewalls and proxies — as services offered by clouds and internet service providers. Sherry joined the CMU faculty in 2016. She earned her master's and Ph.D. in computer science at the University of California, Berkeley, and received the SIGCOMM doctoral dissertation award from the Association for Computing Machinery's Special Interest Group on Data Communications, among other awards.

Crane Receives Packard Fellowship

The David and Lucile Packard Foundation has announced that Keenan Crane, assistant professor of computer science and robotics, is one of 18 recipients of its 2018 Packard Fellowships for Science and Engineering. The fellowship recognizes innovative early-career researchers and includes $875,000 to aid in each fellow's research for five years. Crane's research explores how the shapes and motions observed in nature can be faithfully expressed in a language that is completely finite and discrete, and, hence, be understood by a computer. His exploration of this question provides both new foundations for computation, as well as new ways to turn digital designs into physical, shape-shifting matter. At CMU, his work has included computational tools for translating complex 3D designs into mechanisms that can be built by cutting, bending, inflating or milling physical material, as well as fundamental algorithms for understanding geometric data. Crane joined the CMU faculty in 2015. He earned a Ph.D. in computer science at Caltech and previously completed an NSF Mathematical Sciences Post-Doctoral Fellowship in Columbia University's Department of Computer Science. This year marks the 30th anniversary of the Packard Fellowships. To commemorate the occasion, the foundation has created a new website to celebrate the work, ideas and careers of 30 years of fellows. "Over the past three decades, the fellowship program has been an example of our deep commitment to basic research in science and engineering," said David Orr, chair of the foundation's board of trustees. "The new class will certainly continue the tradition of the groundbreaking science that Packard fellows have become known for."

Sound, Vibration Recognition Boost Context-Aware Computing

New Methods Help Smart Devices Detect What's Happening Around Them

Smart devices can seem dumb if they don't understand where they are or what people around them are doing. Carnegie Mellon University researchers say this environmental awareness can be enhanced by complementary methods for analyzing sound and vibrations."A smart speaker sitting on a kitchen countertop cannot figure out if it is in a kitchen, let alone know what a person is doing in a kitchen," said Chris Harrison, assistant professor in CMU's Human-Computer Interaction Institute (HCII). "But if these devices understood what was happening around them, they could be much more helpful."Harrison and colleagues in the Future Interfaces Group will report today at the Association for Computing Machinery's User Interface Software and Technology Symposium in Berlin about two approaches to this problem — one that uses the most ubiquitous of sensors, the microphone, and another that employs a modern-day version of eavesdropping technology used by the KGB in the 1950s.In the first case, the researchers have sought to develop a sound-based activity recognition system, called Ubicoustics. This system would use the existing microphones in smart speakers, smartphones and smartwatches, enabling them to recognize sounds associated with places, such as bedrooms, kitchens, workshops, entrances and offices."The main idea here is to leverage the professional sound-effect libraries typically used in the entertainment industry," said Gierad Laput, a Ph.D. student in HCII. "They are clean, properly labeled, well-segmented and diverse. Plus, we can transform and project them into hundreds of different variations, creating volumes of data perfect for training deep-learning models."This system could be deployed to an existing device as a software update and work immediately," he added.The plug-and-play system could work in any environment. It could alert the user when someone knocks on the front door, for instance, or move to the next step in a recipe when it detects an activity, such as running a blender or chopping.The researchers, including Karan Ahuja, a Ph.D. student in HCII, and Mayank Goel, assistant professor in the Institute for Software Research, began with an existing model for labeling sounds and tuned it using sound effects from the professional libraries, such as kitchen appliances, power tools, hair dryers, keyboards and other context-specific sounds. They then synthetically altered the sounds to create hundreds of variations.Laput said recognizing sounds and placing them in the correct context is challenging, in part because multiple sounds are often present and can interfere with each other. In their tests, Ubicoustics had an accuracy of about 80 percent — competitive with human accuracy, but not yet good enough to support user applications. Better microphones, higher sampling rates and different model architectures all might increase accuracy with further research.In a separate paper, HCII Ph.D. student Yang Zhang, along with Laput and Harrison, describe what they call Vibrosight, which can detect vibrations in specific locations in a room using laser vibrometry. It is similar to the light-based devices the KGB once used to detect vibrations on reflective surfaces such as windows, allowing them to listen in on the conversations that generated the vibrations."The cool thing about vibration is that it is a byproduct of most human activity," Zhang said. Running on a treadmill, pounding a hammer or typing on a keyboard all create vibrations that can be detected at a distance. "The other cool thing is that vibrations are localized to a surface," he added. Unlike microphones, the vibrations of one activity don't interfere with vibrations from another. And unlike microphones and cameras, monitoring vibrations in specific locations makes this technique discreet and preserves privacy.This method does require a special sensor, a low-power laser combined with a motorized, steerable mirror. The researchers built their experimental device for about $80. Reflective tags — the same material used to make bikes and pedestrians more visible at night — are applied to the objects to be monitored. The sensor can be mounted in a corner of a room and can monitor vibrations for multiple objects.Zhang said the sensor can detect whether a device is on or off with 98 percent accuracy and identify the device with 92 percent accuracy, based on the object's vibration profile. It can also detect movement, such as that of a chair when someone sits in it, and it knows when someone has blocked the sensor's view of a tag, such as when someone is using a sink or an eyewash station.The Packard Foundation, Sloan Foundation and Qualcomm supported the work on Ubicoustics and Vibrosight, with additional funding from the Google Ph.D. Fellowship for Ubicoustics.

Lights, Camera, Science!

SCS Sophomore Abraham Riedel-Mishaan Featured in "Science Fair" Documentary

Abraham Riedel-Mishaan seems an unlikely movie star. He's not an actor. No one stops the sophomore computer science major on the Carnegie Mellon University campus to ask for autographs. But he is featured in "Science Fair," a film festival darling now showing nationwide. The documentary follows Riedel-Mishaan and eight other high school students as they prepare for and compete at the 2017 International Science and Engineering Fair in Los Angeles. Critics have described the film as "immensely likeable," "brilliant and quirky" and an "ode to the teenage science geeks on whom our future depends." "'Science Fair' is a love letter to the subculture that saved me," said Cristina Costantini, who co-directed the film with Darren Foster. "As a dweeby kid growing up in a sports-obsessed high school in Wisconsin, the international science fair became my lifeboat. … Science fair is where I found my tribe." Riedel-Mishaan had already found his tribe by the time Foster, Costantini and director of photography Peter Alton showed up at duPont Manual High School in Louisville, Ky. He was enrolled in the school's acclaimed math and technology magnet program. A summer course in programming robotics sparked an interest in computer science. "I really enjoyed how it was a perfect blend of mathematics and practicality, being able to do computer science both through a mathematical lens and also through the lens of a builder and being able to actually create something," he said. The filmmakers came to duPont Manual because they knew the school regularly sends top competitors to the international fair. In fact, the school has its own regional qualifying fair. "It was cool to have a film crew following us around," Riedel-Mishaan recalled, "but no one really knew what would come of it." The three filmmakers were gathering material on a number of students who had projects that might make it to the international fair. Just who would make the final cut and what would become of the little film was unknown. Inspired by the 200th anniversary of French physician René Laennec's invention of the stethoscope, Riedel-Mishaan and two friends, Ryan Folz and Harsha Paladugu, decided their science project would be an inexpensive, 3D-printable stethoscope that could be used to detect heart disease in medically underserved areas. Paladugu focused on the big picture, including the design and testing of the stethoscope, while Folz concentrated on developing an easy-to-use app and Riedel-Mishaan devised an algorithm that could diagnose heart problems. "Given our use case, we figured it was better to have a false-positive diagnosis rather than to have a false negative," Riedel-Mishaan said, noting that users would be prompted to go to a hospital or a physician for a definitive diagnosis. With that in mind, he selected three previously published diagnostic algorithms and merged them together, weighting the result of each algorithm. "The idea was that the three algorithms together could beat any single existing algorithm," he explained. The team began work on the project in mid-fall 2016, entered their first competition in March 2017 and ultimately qualified for the international fair in May. The Louisville guys had their share of drama at the international fair, at one point being cited for an infraction when one judge mistakenly thought that the team had done human testing. The discrepancy was cleared up. Though the team didn't win any of the major prizes, it did scoop up a number of specialty awards, including one from the Acoustical Society of America. The teammates went their separate ways at graduation — Paladugu to Harvard, Folz to Northeastern and then the University of North Carolina, and Riedel-Mishaan to Carnegie Mellon and the School of Computer Science. "I chose CMU because of how much it embodied that mathematics and theory focus of computer science, and so far I have been really enjoying the ability to dig deeply into theoretical concepts with computer science," Riedel-Mishaan said. The film documentary receded into the background. But then "Science Fair" won the audience award at the Sundance Film Festival and then again at SXSW. And then National Geographic bought the rights. Suddenly, "Science Fair" was no longer a little film. "From there, it became really, really big," he said. Riedel-Mishaan attended the premiere in Los Angeles in September, as did all the featured students except Paladugu, and later participated in a question-and-answer session at the Carnegie Science Center in Pittsburgh, where Science Fair is now showing. Even his friends on campus are beginning to become aware of it, he added. Riedel-Mishaan said he is happy with how the movie turned out. "It's not boring, for sure," he said. "I think it shows a true image of us."

New CMU Degree Prepares Researchers for AI-Directed Experimentation

Artificial Intelligence Will Drive More Decisions in Biological Experiments

Computers increasingly are helping scientists identify and select experiments necessary for scientific discoveries, so Carnegie Mellon University has created a two-year master's degree program to train specialists needed to design, configure, operate and maintain these systems.CMU's Computational Biology Department will offer the Master of Science in Automated Science: Biological Experimentation beginning in fall 2019 and is accepting applications for its initial class through Dec. 1."Automation has disrupted numerous industries and is poised to radically transform the pace and economics of scientific research in academia and industry," said Robert F. Murphy, head of the Computational Biology Department and co-director of the new master's degree program. "We will train students to become leaders in this new field, where automated instruments and artificial intelligence combine to produce scientific discoveries."Automation such as high-throughput screening is a standard means of experimentation for drug discovery and of basic biological science. Advances in AI and machine learning now make it possible and — given the complexity and scope of today's experiments — even preferable for computers to choose which experiments will fill gaps in knowledge and which only duplicate knowledge and can be skipped."The goal is to develop self-driving instruments, similar to self-driving cars that require little if any help from their occupants," Murphy said. "The need for human intervention in experiments will be minimal, though creating these automated systems and planning experiments will require people who are familiar both with experimental methods and with the machine learning and statistical methods necessary to construct predictive models.""This exciting new program in automated science will break new ground while building upon the unique strengths of Carnegie Mellon," said Carnegie Mellon President Farnam Jahanian. "By training a new generation to develop and use 'self-driving instruments' that combine artificial intelligence with automated research instruments, we will play a leading role in advancing new paradigms in discovery and changing the way that experimental science is done."Christopher Langmead, associate professor of computational biology and co-director with Murphy of the master's program, said the initial concentration will be in biological experimentation, but additional subject areas are expected to be added in coming years.The program seeks to attract students who are preparing for laboratory careers and otherwise might pursue master's degrees in biology or chemistry. It will train students for jobs such as laboratory automation specialists and automation engineers. It also will provide excellent preparation for students contemplating Ph.D. studies in related disciplines.To ensure that it meets industry needs, the program will have an external advisory board drawn from potential employers."This program will provide a major boost to the scientific automation field and I am very happy to be involved with it," said DJ Kleinbaum, an external advisory board member and co-founder of Emerald Cloud Lab, a California company that provides automated solutions for contract research.The MSASBE program will provide training in three areas:Hands-on use of automated instruments and study of their design principles, interfaces and capabilities;Computational methods for constructing predictive models from experimental data; andAlgorithmic methods for experimental design and selection.The interdisciplinary program will draw on faculty from CMU's Computational Biology Department, Machine Learning Department, Computer Science Department, Department of Biological Sciences, Department of Chemistry, Mechanical Engineering Department and Biomedical Engineering Department.

Faculty Profile: Comp Bio's Compeau Rethinks Traditional Teaching

When Phillip Compeau was a child, he dreamed of being whisked away to a new kind of school, one where all the students sat at their own desks with a collection of work and a computer to guide them through it. Each student could tackle their personalized task list at their own pace. Need help? Click a button or ask the teacher. It's no surprise, then, that Compeau — now an assistant teaching professor and assistant department head for education in Carnegie Mellon University's Computational Biology Department — has spent much of his professional life studying not just his chosen field of computational biology, but also how students learn. And he's worked hard to find the best strategies for teaching them. In its simplest form, computational biology uses techniques from computer science to solve biological problems. Before computational biology was a mature discipline — CMU awarded its first undergraduate comp bio degrees in 1989, when the program lived in the Mellon College of Science — students interested in the field often majored in computer science and took bio courses, or maybe studied bio with some work in computer science. In Compeau's case, it was math. As an undergrad at Davidson College in his home state of North Carolina, Compeau studied math but worked at its intersection with biology. Specifically, his senior thesis focused on a variant of the pancake-sorting problem: determining the minimum number of spatula flips necessary to transform a stack of differently sized pancakes into a pyramid shape. "If you think of the pancakes as different genes, and the flips as certain large-scale mutations, then pancake-sorting offers a model for transforming one chromosome into another," Compeau said. He also played competitive tennis and at one point had an Association of Tennis Professionals (ATP) ranking in doubles. But that's beside the point. After a brief excursion to Cambridge University, where he earned a master of advanced study degree in mathematics, Compeau headed west to work with noted computational biologist Pavel Pevzner at the University of California, San Diego. While there, his focus changed. "I did research during my Ph.D. years, but the larger effort I undertook was doing online educational projects and building an online textbook," Compeau said. "I decided I wanted to be a teaching-track professor, which is not the most common ambition. I was fortunate enough to be in an environment with an advisor who could support that." While at UCSD, Compeau co-founded Rosalind, an online platform for learning bioinformatics through problem-solving. He also co-instructed the first massive open online course (MOOC) in bioinformatics. Both initiatives have since exploded. Rosalind has reached more than 50,000 people and has been adopted more than 100 times by universities for offline courses. The MOOC has evolved into the bioinformatics specialization on Coursera and its best-selling print companion, "Bioinformatics Algorithms: An Active Learning Approach." The course has been completed by a few thousand learners and used in some way by more than 200,000 people. With Rosalind and the MOOC under his belt, Compeau joined the Computational Biology Department faculty in 2015. "What brought me to CMU was the fact that it has a good history of prioritizing teaching-track faculty and seeing the value in them," he said. "It's a rare environment to have an entire department devoted to computational biology, as well as to be a university that prioritizes teaching." Since arriving in Pittsburgh, Compeau has made substantial contributions to the department. In his role as assistant department head for education, he led the creation of the undergraduate degree in computational biology, which he directs and whose students he advises. It's the only comp bio degree granted from a school of computer science. "Because SCS is so strong at the undergrad level, we can design a major that simply wouldn't be possible at other places," Compeau said. "Students hit the ground running in math and science here, and it means that we can teach them computational biology at a deep level. And that means we can produce graduates who fill a huge area of need for solving the big biological and medical problems of the 21st century." Compeau also collaborated with fellow comp bio teaching professor Josh Kangas to develop the first precollege computational biology program in the U.S. The program, slated to begin this summer, will first put students in the lab and then in front of computers to analyze the data they generate. "It's going to be an amazing experience for them," Compeau said. Like his young self, grown-up Compeau still believes there's massive room for using automation to improve education. In his own courses, he's implemented a flipped classroom and relies on active learning. No traditional lectures here. Students complete reading assignments in an interactive textbook before the class meeting, and spend in-class time working on challenge problems and answering questions from their peers to cement their learning. His strategy, while successful, requires a complete rethinking of learning. And it requires student buy-in. "We indoctrinate students into thinking they can only learn one way," Compeau said. "Teachers are considered the fountain of knowledge, and when that source of wisdom is asking students to solve problems and guiding them to figure out solutions on their own instead. … Well, some people have an allergic reaction to that." To tame that allergy, Compeau spends the first day of his course selling the format to his students by outlining the weaknesses of traditional lectures, the goals of the course and how the flipped course will benefit them. So far, he's seen positive course evaluations and a significant improvement in test scores. While he's made great strides toward building a student-centric classroom, Compeau still hasn't created that fantasy school of his first-grade dreams. But he doesn't want to. "Removing the teacher from the picture entirely is a naïve view,” he said. "The modern classroom should find ways to use online materials to improve our teaching — not to replace it." And at Carnegie Mellon, he's happy to have found a cohort of colleagues willing to do whatever it takes to give students the best education possible. He still, occasionally, plays tennis. These days, though, he's more of a golfer.

Cassell Presents David Award Lecture at National Academies

Justine Cassell, SCS associate dean of technology strategy and impact, will present the 2018 Henry and Bryna David award lecture, "Will Artificial Intelligence Mean the End of Social Interaction?" on Thursday, Oct. 11, at the National Academies of Sciences, Engineering and Medicine in Washington, D.C. The Henry and Bryna David Award honors leading experts for innovative research in the behavioral and social sciences that will affect public policy. Cassell's research focuses on computational systems that use the social bond that arises in conversation and storytelling to enhance people's learning and task performance. She was a keynote speaker at the Grace Hopper Celebration last month and is a past winner of the Anita Borg Institute's Women of Vision Award. She is a fellow of the American Association for the Advancement of Science, the Royal Academy of Scotland and the Association for Computing Machinery. Her award lecture will address fears surrounding the impact of AI on daily lives and work. She will describe some unexpected research results about the ways in which social interaction supports and improves task performance in people. She will also discuss how social interaction can be integrated into AI, with implications for the future of AI, work and social interaction. Her lecture also will be published in Issues in Science and Technology.

New Algorithm Efficiently Finds Antibiotic Candidates

Search Technique Avoids Wasting Time, Expense on Rediscovering Known Compounds

If you're looking for a needle in a haystack, it's best to know what hay looks like. An international team of researchers has applied this idea to the search for new pharmaceuticals, developing a technique that reduces the chances of simply rediscovering known compounds. In an article published today in the journal Nature Communications, researchers from Carnegie Mellon University; the University of California, San Diego; and St. Petersburg State University in Russia describe a new means of searching vast repositories of compounds produced by microbes. By analyzing the mass spectra of the compounds, they were able to identify known compounds within the repository and eliminate them from further analysis, focusing instead on the unknown variants — the needles within the haystack — that might potentially be better or more efficient antibiotics, anticancer drugs or other pharmaceuticals. In just a week, running on 100 computers, the algorithm, called Dereplicator+, sorted through a billion mass spectra in the Global Natural Products Social molecular network at UC San Diego and identified more than 5,000 promising, unknown compounds that merit further investigation, said Hosein Mohimani, assistant professor in CMU's Computational Biology Department and first author on the article. The algorithm that powers this molecular search engine is now available for use by any investigator to study additional repositories. In the past, mass spectrometry data repositories have been underused because it was difficult to search through them and because those efforts to date have been plagued by high rates of rediscovery of known compounds. "It's unbelievable how many times people have rediscovered penicillin," Mohimani said. Analyzing the compounds' mass spectra — essentially, a measurement of the masses within a sample that has been ionized — is a relatively inexpensive way of identifying possible new pharmaceuticals. But existing techniques were largely limited to peptides, which have simple structures such as chains and loops. "We were only looking at the tip of the iceberg," Mohimani said. To analyze the larger number of complex compounds that have entangled structures and numerous loops and branches, the researchers developed a method for predicting how a mass spectrometer would break apart the molecules. Beginning with the weakest rings, the method simulated what would happen as the molecules came apart. Using 5,000 known compounds and their mass spectra, they trained a computer model that could then be used to predict how other compounds would break down. Mohimani said Dereplicator+ not only can identify known compounds that don't need to be investigated further, but it can also find less common variants of the known compounds that likely would go undetected within a sample. In addition to Mohimani, investigators included Alexey Gurevich, Alexander Slemov, Alla Mikheenko, Anton Korobeynikov and Egor Shcherbin of St. Petersburg; Louis-Felix Nothias, Pieter C. Dorrestein and Pavel A. Pevzner of UC San Diego; and Liu Cao of CMU's Computational Biology Department. The National Institutes of Health, Carnegie Mellon and the Russian Science Foundation supported this research.

Mitchell Named CMU's Interim Dean of School of Computer Science

AI and Machine Learning Pioneer To Lead Nation's Top-Ranked Program

Carnegie Mellon University has named Tom Mitchell, the E. Fredkin University Professor of Machine Learning and Computer Science, interim dean of the School of Computer Science. Mitchell is a pioneer in the field of machine learning, a burgeoning branch of artificial intelligence that develops systems capable of learning from data, identifying patterns and making decisions. In 1997, he co-founded the Center for Automated Learning and Discovery, which became the world's first Machine Learning Department in 2006 and offered the first Ph.D. program in machine learning. He led the department until 2016. "As a leading scholar in machine learning and artificial intelligence, Tom Mitchell has been one of the School of Computer Science's most extraordinary founders and pioneers for the past several decades," said Carnegie Mellon University President Farnam Jahanian. "He has the profound respect of the entire Carnegie Mellon community and a record of leadership that will make him an excellent interim dean. I am grateful for his willingness to serve the school and the university at this important time." Mitchell's work has helped to establish Carnegie Mellon's leadership in artificial intelligence research and education. This fall, SCS became the first U.S. university to offer an undergraduate degree in AI. Last year, the university launched its CMU AI initiative to further boost its research efforts and unite AI researchers across the campus. "I have spent 30 years of my career at Carnegie Mellon because there is no better place for generating and exploring new ideas, and for educating the next generation of leaders in computer science and artificial intelligence," Mitchell said. "I look forward to as interim dean working with our faculty, staff and students to move us even further forward, while the university selects our next long-term dean." As interim dean, Mitchell will oversee a school that was named this year the nation's top graduate school of computer science for artificial intelligence by U.S. News and World Report, which once again ranked SCS No. 1 overall, tied with three other computer science schools. Women have reached parity with men in the last three incoming classes of undergraduates, even as the size of first-year classes has increased from 139 in 2014 to a record 211 this year. New K-12 outreach programs are underway, designed to increase the number of underrepresented minorities studying computer science nationwide. Mitchell's research has included the development of statistical learning algorithms and their applications to problems, such as giving computers the ability to understand natural language as well as discovering how the human brain represents information. Mitchell and colleagues in CMU's Psychology Department produced the first computational model to predict brain activation patterns associated with nouns, work that has since been extended to other word types, word sequences and emotions. His Never Ending Language Learner is a computer program that searches through web pages 24/7 as it teaches itself to read. His projects have been featured on CBS's "60 Minutes," PBS's "NOVA Science NOW" and Werner Herzog's 2016 feature documentary, "Lo and Behold." More recently, Mitchell has explored how machine learning, and information technology in general, will affect jobs. He co-chaired a study by the National Academies of Sciences, Engineering, and Medicine that produced a 2017 report on technology and the U.S. workforce. Mitchell earned his undergraduate degree in electrical engineering at Massachusetts Institute of Technology and his Ph.D. in electrical engineering with a minor in computer science at Stanford University. He joined Carnegie Mellon in 1986. He has published and lectured extensively, including at the World Economic Forum's prestigious Davos conference. The university named him the E. Fredkin Professor of AI and Machine Learning in 1999 and a University Professor, CMU's highest faculty distinction, in 2009. A former president of the Association for the Advancement of Artificial Intelligence (AAAI), he is a fellow of both the AAAI and the American Association for the Advancement of Science, and winner of the 2007 AAAI Distinguished Service Award. He was elected in 2010 to the U.S. National Academy of Engineering, and in 2016 to the American Academy of Arts and Sciences. Mitchell will take the place of Andrew Moore, who is stepping down in November at the end of his term as SCS dean to head Google's Cloud AI efforts. Carnegie Mellon will conduct a national search for a successor.