Spring 2016 Colloquia
Check back soon for more information on the computer science seminar series. Unless otherwise noted, the seminars now meet on Mondays at 3pm in Stanley Thomas 302. If you would like to receive notices about upcoming seminars, you can subscribe to the announcement listserv.
A Network-Centric Approach to Data Science: From Distributed Learning to Social Recommender Systems
Zhenming Liu PRINCETON University
Abstract: Networks play important roles in various stages of a data science life cycle, including the design of scalable platforms, the collection of data, and the analysis of statistical models. I will talk about my efforts to develop a suite of network-based techniques in these stages. After briefly describing my work on designing scalable platforms for online machine learning algorithms and that for sampling data from the World Wide Web, I will discuss the details of a recent project that uses network analysis to study social recommender systems. A social recommender system leverages its users’ social connections to improve recommendation service. The recommender system we have designed simultaneously maximizes (a) an individual’s benefit from using a social network and (b) the network’s efficiency in disseminating information. The design solution brings together techniques from spectral analysis, random walk theory, and large-scale optimization.
About the Speaker: Zhenming Liu received his PhD from Harvard University (working with Michael Mitzenmacher) and then spent two years as a postdoc at Princeton University (primarily working with Mung Chiang and Jennifer Rexford). Presently, he is a machine learning researcher for a quantitative hedge fund. Dr. Liu’s research focus is the intersection of data science and network analysis; he designs both algorithms that analyze network structures inherent in the data (e.g., social and biological networks) and scalable platforms in support of big data analytics. He has received several awards for his research, including a Best Paper Runner Up at INFOCOM 2015 and a Best Student Paper Award at ECML/PKDD 2010.
Dérives along Social Ties
Adriana Iamnitchi University of South Florida
This event will be held on Thursday, 2/18/2016, at 2:00 p.m. in Stanley Thomas, Room 302. Please note the special weekday and time for this event.
Abstract: Our digital lives provide a wealth of social information that enables data-driven discoveries of phenomena that cannot be easily observed otherwise. The simple association with other people reveals surprising behaviors at scale.
This talk will present such discoveries in several areas. First, after confirming that the strength of indirect social ties can predict link formation, we show that it also predicts the timing of link formation and how a contagion will propagate. Second, social ties betray unethical behavior online. In particular, we show that content abusers in Yahoo Answers who manage to go under the community radar and thus never have their content “flagged” as inappropriate can be detected from their association with other users in the platform’s social network. And finally, this talk will present how phone conversations—traditional manifestations of social ties—within a mobile cellular network affect the operator's revenues and cost. Plans for future research will also be presented.
About the Speaker: Adriana Iamnitchi is associate professor in the Department of Computer Science and Engineering at University of South Florida. Her research interests are in distributed systems, with current emphasis on designing and evaluating socially-aware distributed systems and on characterizing social networks. Prior to joining USF, Iamnitchi received her PhD from University of Chicago and spent a year at Duke University as visiting assistant professor. Iamnitchi received the US National Science Foundation CAREER Award in 2010.
Health as a Timeline: Machine Learning and Electronic Health Records to Characterize, Predict and Intervene
Jeremy Weiss University of Wisconsin
This event will be held on Monday, 2/22/2016, at 3:30 p.m. in Stanley Thomas, Room 302. Please note the special time for this event.
Abstract: Electronic health records (EHRs) now document 70 percent of medical encounters, a 7-fold increase from a decade ago. Computerization is causing a renovation in health care, and computational techniques are needed to bring the potential of this readily growing data source to full potential in preventing morbidity, educating patients, and encouraging wellness. The forthcoming representation of patient data in EHRs are as timelines, and the field of machine learning is a leader in the development of timeline models. My work contributes to statistical timeline analysis--specifically continuous-time Bayesian networks, point processes, and causal inference--to characterize and predict temporal trajectories. Combining these methods and 40 years of EHR data at the Marshfield Clinic in Wisconsin, my work has shown improvements in making personalized predictions and recommendations for optimal outcomes. Health data and big data frameworks are here now--our next step is to combine these data, frameworks, and analysis to fulfill the promise of improving health outcomes by moving beyond paper.
About the Speaker: Jeremy C Weiss is an MD-PhD candidate at the University of Wisconsin studying machine learning methodology for applications in health care. His work focuses on the development of machine learning methods for the patient-oriented representation of EHR data: timelines. His machine learning contributions in learning and inference increase the scalability of timeline analysis to EHR-scale analyses, and his work on individualizing treatments offers improvements to long-standing clinical analysis frameworks. His research has resulted in publications in NIPS, AMIA, AAAI, and others and has been featured on FiveThirtyEight.com and the Wall Street Journal. He also won Best Talk award at the National Library of Medicine Informatics Training Conference. He is interested in combining his medical expertise with innovations in informatics and computation that will shape the future health of individuals and our communities.
Harnessing Load Elasticity in Networked Systems: Efficiency and Security
Zizhan Zheng University of California at Davis
This event will be held on Wednesday, 2/24/2016, at 3:00 p.m. in Stanley Thomas, Room 302. Please note the special weekday for this event.
Abstract: Many jobs (applications, service requests, etc.) in networked systems exhibit certain level of elasticity. Examples include delay-tolerant applications on the Internet and deferrable electric load in the power grid, as well as interactive applications where a timely result with a good match is preferable to the completed but delayed result. Load elasticity, if properly utilized, can significantly improve the efficiency and scalability of networked systems. An important challenge, however, is that in many real settings, jobs arrive online and decisions have to be made under various types of uncertainty. In this talk, I will discuss my work on exploiting different levels of load elasticity in the design and optimization of networked systems including wireless and sensor networks, cloud systems, and the smart grid, using tools from approximation algorithms, online algorithms, and convex optimization. In particular, I will show that how the increased time elasticity helps achieve more efficient networked systems. However, this is true only when all information is trustworthy. I will then talk about my recent work on data integrity attacks towards the demand-response programs in the smart grid. We find surprisingly that even when the attacker can modify a small fraction of deferrable requests before they are received by the operator, they can actually force a much higher cost than in the case of the “dumb grid” that completely ignores the load elasticity. I will conclude by discussing future research directions, including leveraging online learning to improve the efficiency and security of networked systems, and game theoretic approaches for protecting interdependent systems.
Zizhan Zheng is an associate specialist at the University of California, Davis. From 2010-2014, he was a postdoctoral researcher at The Ohio State University, working with Prof. Ness B. Shroff. Zizhan received his Ph.D. in Computer Science and Engineering in 2010 from The Ohio State University, working with Prof. Prasun Sinha. Prior to joining OSU, he received his B.E. in Polymer Science and Engineering from Sichuan University in China, and his M.S. in Computer Science from Peking University in China. His research interests are in the areas of networking, cloud computing, and cybersecurity.
Building Blocks of the Internet of Things
Jeffrey Voas National Institute of Standards and Technology (NIST)
This colloquium, offered as the 2015-2016 James Mead Lecture (see more info re: Tulane alumnus James Mead ), will be held on Tuesday, March 8th, at 3:30 p.m. in Stanley Thomas 302. Please note the special weekday and time for this event.
Abstract: System primitives allow formalisms, reasoning, simulations, and reliability and security risk-tradeoffs to be formulated and argued. In this work, five core primitives belonging to most distributed systems are presented. These primitives apply well to systems with large amounts of data, scalability concerns, heterogeneity concerns, temporal concerns, and elements of unknown pedigree with possible nefarious intent. These primitives form the basic building blocks for a Network of ‘Things’ (NoT), including the Internet of Things (IoT). This talk discusses the underlying and foundational science of IoT. To our knowledge, the ideas and the manner in which the science underlying IoT is presented here is unique.
About the Speaker: Jeffrey Voas is a computer scientist at the US National Institute of Standards and Technology (NIST) in Gaithersburg, MD. Before joining NIST, Voas co-founded Cigital: www.cigital.com. Voas co-authored two John Wiley books (Software Assessment: Reliability, Safety, and Testability  and Software Fault Injection: Inoculating Software Against Errors . He received two U.S. patents and has over 250 publications. Voas received his undergraduate degree in computer engineering from Tulane University (1985), and received his M.S. and Ph.D. in computer science from the College of William and Mary (1986, 1990 respectively). Voas is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE), the Institution of Engineering and Technology (IET), and the American Association for the Advancement of Science (AAAS). Voas received the U.S. Department of Commerce’s Gold medal in 2014 for his efforts in vetting apps for smartphones for U.S. soldiers in mid-East conflicts. Voas’s current research interests include software certification and the underlying science of IoT. Voas is the Editor-in-Chief of IEEE Transactions on Reliability. Voas is an Adjunct Chair Professor of Computer Science at the National Chiao Tung University in Hsinchu, Taiwan.
Declarative Learning Based Programming for Structured Machine Learning in Natural Language Processing
Parisa Kordjamshidi University of Illinois at Urbana-Champaign
This event will be held on Wednesday, 3/16/2016, at 3:00 p.m. in Stanley Thomas, Room 302. Please note the special weekday for this event.
Abstract: Developing intelligent problem solving systems that deal with real world messy data requires addressing a range of scientific and engineering challenges. Conventional programming languages offer no help to application programmers that attempt to make use of real world data, and reason about it in a way that involves learning interdependent concepts from data, incorporating existing models, and reasoning about them. Over the last few years the research community has tried to address these problems from multiple perspectives, most notably various approaches based on Probabilistic programming, Logical programming and integrated paradigms. In this talk I present Saul, a new declarative learning based programming (DeLBP) language that aims at facilitating the design and development of intelligent real world applications that use machine learning and reasoning. Our new language addresses the following challenges: Interaction with messy data; Specifying the problem at a high level i.e. application level; Dealing with uncertainty in data and knowledge; Supporting structured learning and reasoning while considering expert knowledge that is represented declaratively. An additional advantage of such a paradigm is generating easily reusable models and code, henceforth increasing the replicability of research results. I exemplify the flexibility and the expressive power of this language using a number of applications in natural language processing domain.
About the Speaker: Parisa Kordjamshidi is a postdoctoral researcher in University of Illinois at Urbana-Champaign, CS department, in cognitive computation group. She obtained her PhD degree from KULeuven in July 2013. Her current research foci is on declarative learning based programming (DeLBP). The goal of this programming paradigm is to facilitate programming for building systems that require a number of learning and reasoning components that interact with each other. Such a language would help experts in various domains who are not experts in machine learning, to design complex intelligent systems. During her PhD research on machine learning and natural language processing she introduced the new problem of “Spatial Role Labeling” with the goal of bridging the gap between natural language and formal spatial representation and reasoning models. She provided the first machine learning benchmark for this task. She has investigated structured output prediction and relational learning models to map natural language onto formal spatial representations, appropriate for spatial reasoning as well as to extract knowledge from biomedical text. She is currently involved in an NIH (National Institute of Health) project, extending her research experience on structured and relational learning for biological data analysis. The results of her research have been published in several international peer-reviewed conferences and journals including ACM-TSLP, Journal of Web Semantics, BMC-Bioinformatics and IJCAI.