Over 10,000 have uploaded their genome data to DNA.Land, a crowd-sourcing website that taps genome info for scientific research. The site was launched
in October by CS professor Yaniv Erlich and Joe Pickrell.
Month: December 2015
Dragomir Radev, PhD’99 and visiting professor at Columbia, is elected ACM Fellow
For “contributions to natural language processing and computational linguistics.”
Computer Science Department adds two lecturers to faculty
The number of students seeking a computer science degree is surging. The trend is nationwide and especially strong at Columbia, which this year saw a 30% increase over last year in the number of undergraduates majoring in computer science. This year’s increase follows several years of double-digit increases, including a 50% jump three years ago.
“We’re lucky to have great student interest in computer science, but this also poses challenges. Each of our students needs to be taught the fundamental concepts underlying computer science so that they not only have a strong foundation, but also have the skills to go out and find new ways and new contexts to apply those skills. We need faculty concentrated on this important task,” says Rocco Servedio, chair of the Computer Science department. “The department is fortunate to have Paul Blaer and Ansaf Salleb-Aouissi. Both are extremely knowledgeable in their respective areas of computer science, and just as importantly, both have proven teaching abilities. They are an important addition to the department.”
Columbia this fall promoted Paul Blaer from adjunct professor to Lecturer in Discipline, a full-time faculty position that makes teaching Blaer’s primary focus, something he’s wanted for a long time.
Hiring Blaer full time is not exactly a stab in the dark for Columbia where Blaer is a well-known quantity. Since he was 3, he has been floating around campus. His father is physics professor Allan Blaer, who did both undergraduate and graduate work at Columbia and who—after teaching stints at Princeton and Swarthmore—returned to Columbia where his son would likewise attend as both undergraduate and graduate student.
Paul Blaer did his PhD research in the area of mobile robotics and 3D vision, working in Peter Allen’s lab. It was there, while a grad student leading recitations, that he got his first taste of teaching. He knew immediately that teaching was what he wanted to do. For him, it was the fun stuff, a chance to engage with students, to think on his feet to get them work through problems themselves. For three years before graduating, he was a preceptor, running classes and seeing results from the front of a class.
With teaching in mind, Blaer originally planned to seek a position at a small four-year college, but the combined draw of Columbia and New York City proved strong, and Blaer, knowing Columbia as well as he did, figured someday a teaching opportunity would open up. Until it did, there were other ways on campus for him to contribute.
In Allen’s lab Blaer had been doing systems work—his skills for controlling his own computing environments scaled up for a lab of 50 or more—which led to a full-time position at Computing Research Facilities (CRF); precepting led to part-time adjuncting. For seven years now, Blaer has been teaching introductory computer science classes part-time while working at CRF full time to help faculty design and build backend systems for all types of research projects.
With his new position, the mix gets recalibrated: teaching becomes full-time and CRF part-time.
“I’m thrilled to be working full time with students here at Columbia. It’s the best of both worlds: a large university environment with highly motivated students, yet like a college professor I have this direct interaction with the students, which is the favorite part of my job.”
Blaer knows the classes, the students and faculty, the projects, and how the computer systems are set up; in a department dependent on systems, that’s better than knowing where the bodies are buried. He’s involved also in the administrative aspects that touch on teaching; he is Director of Undergraduate Studies for BS Programs and is active in the Science Honors Program for area high-school science and math students.
Deep institutional and systems knowledge is all well and good, but a lecturer first and foremost has to be able to teach. Blaer has that angle covered especially well. As someone who genuinely cares about teaching, he pays attention to what resonates with students and what doesn’t, and strives to keep his lectures engaging, using humor and real-life stories from his own research to keep students interested. That he succeeds is clear from student comments on the Columbia Underground Listing on Teacher Abilities (CULPA) site, where Blaer has earned a silver nugget for his teaching and approachability.
“We’re thrilled to have Paul join the faculty full-time as a lecturer. The department has rock-solid confidence in his classroom skills because we have the strongest possible kind of evidence—actual results over several years,” says Rocco Servedio, chair of the Computer Science department.
Ansaf Salleb-Aouissi
Introduction to Data Science, Machine Learning for Data Science, Discrete Math, Artificial Intelligence
The increasingly data-centric approach in all aspects of science and technology means students need to learn what algorithms and methods can stand up to the immense scale of today’s data sets. Teaching computer science from the perspective of large data sets is the job of Ansaf Salleb-Aouissi. A data scientist from before the term was commonly understood, Salleb-Aouissi has worked with all types of data on projects ranging from geology and geographic information systems early in her career, to social sciences and urban design, and more recently to medical informatics and to education.
“The common denominator is data. The context may be different and the goals may be different, but at the end of the day, data is data and you try to leverage that data somehow to learn something new,” says Salleb-Aouissi.
An associate research scientist at Columbia’s Center for Computational Learning Systems (CCLS) since 2006, she has worked on both fundamental research into new machine learning and data mining algorithms and methods as well as real-world applications of those methods.
Many of her projects are predictive in nature, forecasting when power-grid failures are likely to occur in one case, and in another predicting which expectant mothers are most, or least, likely to deliver preterm. In this last example, Salleb-Aouissi, with support from the National Science Foundation Smart and Connected Health program, used advanced machine-learning methods to vastly expand the number of risk factors to be considered, including socioeconomic, psychological and behavioral factors.
Prediction is also at the heart of her most recent (and current favorite) project: a web browser optimized for self-learning. “We want to create a personalized self-learning experience by sifting through huge number of search results to identify and return those customized for student’s learning preferences—whether they be videos, books, blogs—and that also fit within the student’s short or long time constraints. The challenge here, as it was with the preterm study, is making all these different and heterogeneous resources work together in a system. It’s an ambitious project and I am very excited to work on it. More so because it is a link between my research and my teaching.”
Though research forms the bulk of her recent work, teaching has also been a component. Post-PhD, she worked as an adjunct professor at the University of Orléans and discovered how much she enjoyed interacting with students. She would have gladly accepted the position of assistant professor except for her plans to eventually move to the US. Instead she took the offer of a Postdoctoral Fellowship at the prestigious research lab INRIA (French National Institute of Computer Science and Control) at Rennes, France. There she did more fundamental investigation of new algorithms, particularly new methods for quantitative association rules, but also for frequent patterns matching, ranking, characterization, and action recommendation.
While still at INRIA, she applied to the CCLS for an open position. Though that position filled quickly, David Waltz, then director of the CCLS, took note of her INRIA fellowship and her growing publications list and contacted her when a different position came up. She and Waltz later collaborated on a number of papers and projects. “Dave smoothed my transition to the CCLS and helped make it an enriching experience where I could grow and learn. I will always be grateful to him.”
Once settled in at the CCLS, she was able to get back into teaching, adjuncting in the Computer Science department, teaching courses in data science, discrete math, and artificial intelligence. As a lecturer, teaching will now be her primary focus, but she will continue doing research, which will now serve a double purpose. “I like to deliver my lecture in an engaging and interactive way, my own way and to keep the material fresh and alive so students actively absorb it rather than just be passive recipients. My own research may serve to give students a peek into what you can do with computer science, and I hope that can motivate them and spark their interest so they learn now so they can do later.”
Right time, right place: A collaborative approach for accurate context-awareness in mobile apps and ads
The right information delivered at the right time can make apps and ads more appealing and relevant to customers: a traffic app that auto-updates for the work or home commute as appropriate; a restaurant that offers lunch coupons for people who work in the area but dinner coupons for people who live nearby. This level of customization requires taking into account a user’s immediate context, something that is not easy to do. It requires both location data and a temporal framework that gives meaning to each location, identifying it as home, work, commute, or another place frequented by a user. But location data is often surprisingly sparse for any one user. To overcome sparsity and construct reliable weekly routines for individual users, researchers integrated global temporal patterns inferred from the entire data set with user-specific spatiotemporal data. The resulting method is entirely data-driven—requiring no labeling—and flexible to accommodate variations in a user’s weekly schedule.
The location data needed for context-aware ads and apps is surprisingly sparse for any single user. For privacy and to conserve energy, most smart phone apps log users’ locations only when the app is active. The result is that location data sets collected from apps comprise many users but few observations per user. This sparsity makes it difficult to know how a particular GPS position is relevant to a user, whether it represents a work place, home, a point along the morning or evening commute, or some other frequently visited destination. It’s this contextual information that allows companies to customize their apps and ads for their customers’ immediate or near-future locations. Local businesses especially benefit when they can accurately predict who is or will soon be in their area.
(It’s not just companies that want context-aware apps and ads; there is evidence users do, too. Cisco found that half of customers surveyed would use coupons sent from a nearby store.)
Location data for any single user may be too sparse to understand when a user transitions between places, but the collection of data across all users represents much more information that can help illuminate broader patterns. To exploit this collective information, four researchers—Berk Kapicioglu, David S. Rosenberg, Robert Schapire, and Tony Jebara—developed a data-driven method that learns people’s important places based on global temporal patterns inferred from the entire data set. They described this method in the paper Collaborative Place Models presented in July at the International Joint Conference on Artificial Intelligence.
Collaborative place models differ from previous methods that first label locations according to time of day and day of week. By assuming, for example, a 9-to-5 workday Monday through Friday, methods that rely on labeling might average positions between 8am and 6pm and call that home while averaging positions between 9 and 5 and calling that work. It’s an intuitive approach but it lacks flexibility—not everyone has the same schedule—and it ignores the commute, which can be a significant amount of time for some people and a missed opportunity for those businesses located along the commute.
Rather than imposing a static temporal framework, collaborative place models learn the quantitative relationship between week-hours by inferring similarities across all users, relying on Bayesian estimation techniques to do so. With a global temporal framework thus set, the relevance of the sparser latitude-longitude GPS coordinates from individual users can then be determined from how they fit into the global temporal pattern. In this way, the model re-constructs a particular user’s home-work-commuting schedule even though a user might have been observed only at Thursday 3pm and Monday 1pm.
To prove the concept, the researchers tested the model using two real-world data sets, a sparse one collected from a mobile ad exchange, and a dense data set from a cellular carrier. In both cases, the only inputs were user IDs, latitudes, longitudes, and time stamps. (Data was anonymized by removing all personal information.)
With data aggregated across all users, a strong, global temporal pattern emerged fairly quickly, one that contained within it several temporal clusters correlated with work, morning and evening commutes, leisure times after work, and sleeping at night. With the global pattern thus established, the individual spatiotemporal patterns of individual users became apparent even with few data points associated with each user.
The spatial extent of place types associated with temporal clusters were determined by replacing multiple observations logged during the same hour with their geometric median (computed using Weiszfeld’s algorithm and by clustering nearby points using a Gaussian mixture model that is a subcomponent of the collaborative place model). This contrasts with the use of averaging in other place models to handle redundant observations and the noise that occurs from GPS errors and from having multiple cell towers covering the same location; by not averaging, the collaborative place model avoids the strange results sometimes caused by deviations in the regular routine, such as a late work evening or a night or weekend away from home.
Flexibility was built into the model by allowing users to have varying numbers of places or week hours. This flexibility turned out to be key; an early, simpler prototype that constrained users to have the same week-hour distribution performed worse than a baseline model.
In the end, data by itself was enough to reliably assess a user’s spatiotemporal schedule. Without the need to label or average location places, the collaborative approach of combining global patterns with sparse user location data reduced the median distance error by 8% from a simpler non-collaborative baseline model.
-Linda Crane
Posted 12/1/2015
About the researchers
Tony Jebara is Associate Professor of Computer Science at Columbia University where he directs the Columbia Machine Learning Laboratory. His research intersects computer science and statistics to develop new frameworks for learning from data with applications in vision, networks, spatiotemporal data, and text. Jebara has founded and advised several startups including Sense Networks (acquired by YP in 2014), Agolo, Ninoh and Bookt. He has published over 100 peer-reviewed papers in conferences, workshops, and journals and is the author of Machine Learning: Discriminative and Generative and co-inventor on multiple patents in vision, learning and spatiotemporal modeling. He is the recipient of the Career award from the National Science Foundation (NSF) and has received honors from the International Conference on Machine Learning and from the Pattern Recognition Society. Jebara’s research has been featured on television (ABC, BBC, New York One, TechTV, etc.) as well as in the popular press (New York Times, Slash Dot, Wired, Businessweek, IEEE Spectrum, etc.). He obtained his PhD in 2002 from MIT.
Berk Kapicioglu is a data scientist at Foursquare and was previously at Sense Networks, a location analytics firm where he worked on location prediction, place modeling, venue recommendation, and real-time ad targeting. He obtained his PhD and MA degrees in computer science department at Princeton University. His thesis focused on the design and analysis of machine learning algorithms for spatiotemporal datasets, and he was advised by Robert Schapire. His undergraduate degree is from the University of Pennsylvania, where he triple-majored in computer science, math, and philosophy.
David S. Rosenberg is a data scientist within the CTO Office at Bloomberg LP. Previously he was Chief Scientist at both YP Mobile Labs and at Sense Networks. Rosenberg specializes in machine learning, artificial intelligence, predictive analytics and statistical modeling, and earned his Bachelor of Science degree in mathematics from Yale University, his Master of Science degree in applied math (computer science focus) from Harvard University, and his PhD in statistics from the University of California, Berkeley. He holds four US patents.
Robert Schapire is a Principal Researcher at Microsoft Research in New York City. He received his PhD from MIT in 1991. After a short post-doc at Harvard, he joined the technical staff at AT&T Labs (formerly AT&T Bell Laboratories) in 1991. In 2002, he became a Professor of Computer Science at Princeton University and joined Microsoft Research in 2014. His awards include the 1991 ACM Doctoral Dissertation Award, the 2003 Gödel Prize, and the 2004 Kanelakkis Theory and Practice Award (both of the last two with Yoav Freund). He is a fellow of the AAAI, and a member of the National Academy of Engineering. His main research interest is in theoretical and applied machine learning, with particular focus on boosting, online learning, game theory, and maximum entropy.
Add to GMail