STATISTICS ON NETWORKS WORKSHOP

September 26-27, 2005

ABSTRACTS AND SPEAKERS’ BIOS

Abstract: A surprisingly consistent view on the fundamental nature of complex systems can now be drawn from the convergence of three distinct research themes. First, molecular biology has provided a detailed description of much of the components of biological networks, and the organizational principles of these networks are becoming increasingly apparent. It is now clear that much of the complexity in biology is driven by its regulatory networks, however poorly understood the details remain. Second, advanced technology is creating engineering examples of networks where we do know all the details and that have complexity approaching that of biology. While the components are entirely different, there is striking convergence at the network level of the architecture and the role of protocols, layering, control, and feedback in structuring complex system modularity. Finally, there is a new mathematical framework for the study of complex networks that suggests that this apparent network-level evolutionary convergence both within biology and between biology and technology is not accidental, and follows necessarily from the shared requirements that systems be efficient, robust, adaptive, and evolvable. A crucial insight is that both evolution and natural selection or engineering design must produce high robustness to uncertain environments and components in order for systems to persist. Yet this allows and even facilitates severe fragility to novel perturbations, particularly those that exploit the very mechanisms providing robustness, and this “robust yet fragile” (RYF) feature must be exploited explicitly in any theory that hopes to scale to large systems. This view of “organized complexity” in biology, technology, and mathematics contrasts sharply with the view of “emergent complexity” popular in other areas of science, and this can now be made mathematically precise. This talk will focus on motivating examples from biology, technology, ecosystems, and large scale natural and technological disasters, and the statistical issues that arise in rigorously studying them.

Bio: John Doyle is the John G Braun Professor of Control and Dynamical Systems, Electrical Engineer, and BioEngineering at Caltech. He has a BS and MS in EE, MIT (1977), and a PhD, Math, UC Berkeley (1984). Early work was in the mathematics of robust control, LQG robustness, (structured) singular value analysis, H-infinity and many recent extensions. He coauthored several books and software toolboxes currently used at over 1,000 sites worldwide, the main control analysis tool for high performance commercial and military aerospace systems, as well as many other industrial systems. Early example industrial applications include X-29, F-16XL, F-15 SMTP, B-1, B-2, 757, Shuttle Orbiter, electric power generation, distillation, catalytic reactors, backhoe slope-finishing, active suspension, and CD players. Current research interests are in theoretical foundations for complex networks in engineering and biology, as well as multiscale physics. His group led the development of the open source Systems Biology Markup Language (SBML) and the Systems Biology Workbench (SBW), which have become the central software infrastructures for systems biology (www.sbml.org), and also released the analysis toolbox SOSTOOLS (www.cds.caltech.edu/sostools). He was the theoretical lead on the team that developed the FAST protocol and shattered multiple world land speed records (netlab.caltech.edu). Prize papers include the IEEE Baker (for the top research paper in all of the IEEE’s ~90 journals, also ranked in the top 10 “most important” papers world-wide in pure and applied mathematics from 1981-1993), the IEEE Automatic Control Transactions Axelby (twice), and the AACC Schuck. Individual awards include the IEEE Control Systems Field Award (2004) and the Centennial Outstanding Young Engineer (1984). He has held national and world records and championships in various sports.

Abstract: Much of computational neuroscience is based on the Hodgkin-Huxley equations, which describe electrical activity of individual neurons and networks of them. The talk will start with a description of these equations and the different ways in which networks are formed. Examples are taken from networks that produce the neural rhythms seen in various cognitive states. These examples focus on statistical issues that arise in modeling neural dynamics.

Bio: Nancy Kopell has a Ph.D. in mathematics, and has been working in neuroscience for a decade or two. Her mathematical focus is dynamical systems, especially geometrical theory of systems with multiple time scales. Scientifically, she has worked on a range of questions including pattern formation in chemical systems and central pattern generators for locomotion. She is currently obsessed with how the nervous system makes use of its dynamics, especially its rhythmic dynamics, to help with sensory processing, cognition and motor preparation.

Abstract: The simplest models of networks treat vertices as indistinguishable, having no characteristics other than their network position. In most real-world networks, however, vertices possess a variety of distinguishing features, such as geographical location or age for individuals in social networks, or textual content for web pages. Moreover, the characteristics of network vertices are known in many cases to have a large effect on patterns of network connections. This talk will focus on measures and models of assortative and disassortative mixing -- the tendency for vertices to connect to others that are like or unlike them in some respect -- and the effect that such mixing has on large-scale network structure, particularly the formation of "communities." Some recent progress on graph partitioning methods will also be described, along with results showing how such methods can be usefully applied to the detection of community structure in networks.

Bio: Mark Newman received his PhD in theoretical physics from the University of Oxford in 1991 and worked at Cornell University and the Santa Fe Institute before moving to the University of Michigan in 2002. He is currently Associate Professor of Physics and Complex Systems at the University of Michigan and a member of the external faculty of the Santa Fe Institute. He has research interest in network statistics and modeling, epidemiology, computer algorithms, and cartography.

Abstract: Some recently developed statistical models for social networks embed the actors in an unobserved "latent social space," which can be used as a parsimonious description of complicated patterns in the network structure. One outstanding issue has been the determination of the dimension of this latent space. In this talk we show how Bayesian methods can be used to select an appropriate dimension, and how Bayesian model averaging over the dimension can improve upon the predictive power of social network models. We illustrate this with an example in international relations, modeling conflicts among nations during 1990-2000.

Bio: Peter Hoff is an assistant professor in the departments of Statistics and Biostatistics, and a member of the Center for Statistics and the Social Sciences at the University of Washington in Seattle.

Abstract: Embedded Networked Sensing represents an exciting class of computing systems that combine distributed sensing, computation, actuation and wireless communication. On the technology side these systems are being touted as a technology as disruptive and enabling as the Internet; on the application side they are currently being used to study terrestrial and aquatic ecosystems and geochemical cycles, and will ultimately be used to monitor public exposure to contaminants, manage land use, and support safer structures. I will discuss experiences with early prototypes and suggest directions for future research, including the importance of supporting interactive as well as autonomous system features.

Bio: Deborah Estrin (Ph.D. MIT EECS (1985), B.S. UCB EECS (1980)) is a Professor of Computer Science at UCLA where she holds the Jon Postel Chair in Computer Networks and is the founding Director of the NSF Science and Technology Center for Embedded Networked Sensing (CENS). Estrin has been instrumental in defining the research agenda for wireless sensor networks. Estrin's research focuses on technical challenges posed by autonomous, distributed, physically coupled systems. She is particularly interested in environmental monitoring applications and is on the National Ecological Observatory Network (NEON) design team. Earlier in her career Estrin contributed to the design of Internet routing protocols. Estrin is a member of the NSF CISE Advisory Committee and the NRC CSTB.

Abstract: Information processing in mammalian cells occurs by interactions between cellular components. The components and links make up a network of coupled chemical reactions. Such networks have substantial information processing ability. To obtain an initial overview of the dynamic topology of such a network and how it can process information, we developed an in silico system of 545 components (nodes) and 1259 interactions representing signaling pathways and cellular machines in the hippocampal CA1 neuron. Using graph theory methods, we analyzed signal flow through the system induced by ligand-receptor interactions. Networking resulted in the emergence of regulatory motifs, such as positive and negative feedback and feedforward loops and bifan motifs that process information. Key regulators of neuronal plasticity were highly connected nodes within the network and these were required for the formation of the regulatory motifs. Specification of input and output nodes allowed us to identify functional modules within the network and the regulatory balance of regulatory motifs within these modules. These data suggest that such a balance of regulatory motifs may be important in determining cellular choices between homeostasis and plasticity.

Bio: Ravi Iyengar is the Rosenstiel Professor and Chair of the Department of Pharmacology and Biological Chemistry at Mount Sinai School of Medicine, New York NY. Trained as a biochemist, Dr Iyengar has used biochemical and molecular biological approaches to study cell signaling, with a focus on heterotrimeric G protein pathways. Over the past decade the Iyengar laboratory has also used computational approaches to understand the regulatory capabilities of cellular signaling networks. Their most recent studies using graph theory approaches was published in Science in August 2005.

Abstract: Dynamic network analysis (DNA) is an emergent field centered on the collection, analysis, understanding and prediction of dynamic relations (such as who talks to whom and who knows what) and the impact of such dynamics on individual and group behavior. DNA facilitates reasoning about real groups as complex dynamic systems that evolve over time. In this talk, a basic tool chain for DNA is described and then their use is illustrated by examining al Qaeda. Technology described enables the analyst to identify vulnerabilities in the terrorist network and to assess how that network might change in response to strategic interventions.

Bio: Kathleen M. Carley received her Ph.D. from Harvard in mathematical Sociology. She is currently a professor of computer science at Carnegie Mellon University. She also directs the center for Computational Analysis of Social and Organizational Systems (CASOS). CASOS is a university wide center for understanding complex systems through the combined application of computer science, social science and social networks. Her research combines cognitive science, social networks and computer science to address complex social and organizational problems. Her specific research areas are computational social and organization theory, group, organizational and social adaptation and evolution, social and dynamic network analysis, computational text analysis, and the impact of telecommunication technologies and policy on communication, information diffusion, disease contagion and response within and among groups particularly in disaster or crisis situations. Her models meld multi-agent technology with network dynamics and empirical data. Three of the large-scale multi-agent network models she and the CASOS group have developed in the counter-terrorism area are:

Abstract: Over the past ten years, we have tested and helped develop a multi-electrode array for chronic cortical recordings in behaving non-human primates. We have found that it is feasible to record from dozens of single units in the motor cortex for extended periods of time and that these signals can be decoded in a closed-loop, real-time system to generate goal-directed behavior of external devices. This work has culminated in a FDA clinical trial that has demonstrated that a tetraplegic patient can voluntarily modulate motor cortical activity in order to move a computer cursor to visual targets. Further advances in BMI technology using non-human primates have focused on using multiple modes of control from signals in different cortical areas. We demonstrate that primary motor cortical activity may be optimized for continuous movement control whereas signals from the premotor cortex may be better suited for discrete target selection. We propose a hybrid BMI whereby decoding can be voluntarily switched from discrete to continuous control modes.

Bio: Dr. Hatsopoulos received his B.A. in physics in 1984 from Williams College. He received a masters (Sc.M.) in Psychology in 1991 and a Ph.D. in Cognitive Science in 1992 both from Brown University. He was a postdoctoral fellow at Caltech from 1992-1995 and then again at Brown University from 1995--1998. From 1998-2001, he was an Assistant Professor of Research in the Dept. of Neuroscience at Brown University. From 2002 to the present, he has been an Assistant Professor in the Dept. of Organismal Biology and Anatomy and in the Committees of Computational Neuroscience and Neurobiology at the University of Chicago.

Abstract: It is understood that, generally speaking, the statistical analysis of data can be impacted in fundamental ways by the manner in which the data were obtained. That is, sampling design and methodology can have important, and sometimes subtle, effects on statistical inferences. The current proliferation of networks and network analysis across the sciences brings with it new challenges on the topic of sampling. One such challenge is that of drawing inferences from path-based sampling in the Internet. I will present recent work looking at this issue in two contexts: (i) inference of Internet attributes, and (ii) inference of global network traffic summaries. In both cases, one finds that there are important interactions between network structure and inferential accuracy.

Bio: Professor Kolaczyk is Associate Professor of Statistics and Director of the Program in Statistics, in the Department of Mathematics and Statistics, and member of the Center for Information and Systems Engineering (CISE), at Boston University. His research focuses on the statistical modeling and analysis of various types of temporal, spatial, and network data, with a particular emphasis on the use of sparseness in inference. His work has resulted in new methods for signal and image denoising, tomographic image reconstruction, disease mapping, high-level image analysis in land cover classification, and analysis of computer network measurements. Professor Kolaczyk's publications have appeared in the literatures on statistical theory and methods, engineering, astronomy, geography, and computer science. His work has been supported by various grants from the National Science Foundation and the Office of Naval Research.

Abstract: An historical overview of the field of social network analysis. I attempt to outline the key elements that distinguish the paradigm with respect to theory, methodology and data. In addition, I attempt to critically assess the field in terms of classical criticisms, such as lack of theory, neglect of agency, neglect of dynamics, etc.

Bio: Professor Borgatti received his Ph.D. in Mathematical Social Science from the University of California, Irvine in 1989. His research interests are in shared cognition and social networks. He is the author of Anthropac, a software package for cultural domain analysis, and UCINET, a software package for social network analysis. He is a past director of the NSF Summer Institute for Research Methods, as well as a past president of INSNA, the professional association for social network researchers. He currently serves as associate editor for a number of journals including Field Methods and Computational and Mathematical Organizational Theory, as well as senior editor for Organization Science. He is currently professor and chair of the Organization Studies department at the Carroll School of Management at Boston College.

Abstract: In the cerebral cortex, the vasculature forms a highly interconnected 2-D networks of surface vessel that supplies blood to different columns, as well 3-D networks of subsurface microvessels within cortical columns. Here I discuss the dynamics of flow in these networks and contrast the effects of perturbations to the flow in the surface arterioles as compared to the network of subsurface capillaries. The impact of these data on blood flow in the normal and diseased brain will be discussed, along with the relationship between flow and topology.

Bio: David, a native of Brooklyn who miraculously finds himself living in La Jolla, is part of a generation of scientists who trained in Physics in the 80's and 90's and presently devote themselves to problems in the Neurosciences. David focuses on feedback control in somatosensation - using the rat vibrissa sensorimotor system as a model - and on blood flow dynamics in vascular loops - using rodent neocortex as a model system. Aspects of the later work involve the use of nonlinear optics as a tool to measure and perturb flow. David takes particular pride in the advanced education of graduate and postdoctoral students through his involvement in summer programs - on Computational Modeling, Data Analysis, and Imaging - held at Woods Hole and at Cold Spring Harbor.

Abstract: I present an overview of natural disasters (wildfires, earthquakes, tsunamis, and hurricanes), describing both the physical phenomena and the social and economic impact. I highlight recent events in the context of long term planetary processes (e.g. plate tectonics). All of the disaster phenomena exhibit power law distributions in the frequency of events as a function of size, so that while most events (e.g. the median size) are small, the losses are dominated by the relatively few largest events. Disaster phenomena are an intrinsic part of ecosystem dynamics, and are important for sustaining diversity. Recent event sizes, while dramatic, are in most cases consistent with long term statistics. However, the associated social and economic costs are clearly on the rise, and this can be attributed to increasing human populations and urbanization, especially in coastal areas, which are especially prone to hydrological and geophysical activity. Effective strategies for mitigating hazards must account for resource constraints across a broad range of scales, and heterogeneous event types, and utilize proactive modeling, education, and planning strategies. Cascading failures, which propagate from the physical phenomena to social and technological infrastructure, impact transportation and communication systems, as well as political and economic conditions on a global scale.

Bio: Jean M. Carlson received a B.S.E. in Electrical Engineering and Computer Science from Princeton University in 1984, an M.S.E. in Applied and Engineering Physics from Cornell University in 1987, and a Ph.D. in Theoretical Condensed Matter Physics from Cornell in 1988. After Postdoctoral work at the Institute for Theoretical Physics, at the University of California Santa Barbara, she joined the faculty at UCSB in 1990, where she is currently a Professor of Physics. She is a recipient of Fellowship Awards from the Sloan Foundation, the David and Lucile Packard Foundation, and the McDonnell Foundation. Carlson's research interests include a combination of foundational work and

a variety of practical applications of complex systems theory, including earthquakes, wildfires, and optimization and design in networks.

Abstract: Exponential family random graph models attempt to represent the complex dependencies in networks in a parsimonious, tractable and interpretable way. A major barrier to the application such models has been lack of understanding of model behavior and a sound statistical theory to evaluate model fit. This problem has at least three aspects: the specification of realistic models; the algorithmic difficulties of the inferential methods; and the assessment of the degree to which the graph structure produced by the models matches that of the data. In this talk we address these through the ideas of degeneracy and stability for some commonly used models.

Bio: Mark S. Handcock is a Professor of Statistics and Sociology, Department of Statistics, University of Washington, Seattle. His work focuses on the development of statistical models for the analysis of social network data, spatial processes and demography. He received his B.Sc. from the University of Western Australia and his Ph.D. from the University of Chicago. Descriptions of his work is available at http://www.stat.washington.edu/handcock.

Abstract: Commonly used interdependency measures such as cross correlation and spectral coherence do not yield directional information. Phase spectra may be used for that purpose only under very ideal conditions. Recent work has begun to explore the use of causal measures to further dissect the interaction patterns among neural signals. In this talk I will describe the concept of Granger Causality and introduce Geweke’s causality spectra. The technique will then be applied to the analysis of multichannel local field potentials recorded from behaving monkeys performing sensorimotor and selective attention tasks.

Bio: Mingzhou Ding received his BS in Astrophysics from Peking University in 1982 and his PhD in physics from the University of Maryland in 1990. He is currently a professor in the Department of Biomedical Engineering at the University of Florida. His main research interest includes cognitive neuroscience and related signal processing problems.

Abstract: The study of complex networks has emerged over the past several years as a theme that spans many disciplines, ranging from computing and information science to the social and biological sciences. Indeed, a shared interest in network structure is arguably one of the forces that is helping draw many of these disciplines closer together. One of the challenges in studying large-scale networks is the extent to which they become intertwined with one another and with their environments. Social and technological networks are becoming increasingly dependent on one another, and both types of networks are fundamentally embedded in time, space, and organizational structure. I'll discuss how these issues arise in the context of search problems and `small-world' effects in networks, focusing on some basic models that capture the notion of a network embedded in spatial and organizational structure, and the surprising extent to which these models are reflected in recent experimental data. I'll also discuss some recent studies of how large-scale networks evolve over time, including a few findings that suggest the limitations of our current models.

Bio: Jon Kleinberg received his Ph.D. in computer science from MIT in 1996; he subsequently spent a year as a Visiting Scientist at the IBM Almaden Research Center, and is now a Professor in the Department of Computer Science at Cornell University. His research interests are centered around issues at the interface of networks and information, with an emphasis on the social and information networks that underpin the Web and other on-line media. He is the recipient of an NSF Career Award, an ONR Young Investigator Award, an Alfred P. Sloan Foundation Fellowship, a David and Lucile Packard Foundation Fellowship, teaching awards from the Cornell Engineering College and Computer Science Department, and the 2001 National Academy of Sciences Award for Initiatives in Research.

POSTER PRESENTATIONS

Abstract: Understanding Internet topology is critical for a variety of pressing networking issues, ranging from evaluating the performance of new routing protocols to assessing the effectiveness of proposed security measures intended to protect the network from nefarious intrusions and attacks. However, since little is known about the relationship between large-scale statistical properties of a network topology and its underlying structural characteristics, there continues to be significant confusion and controversy surrounding the area of Internet topology modeling, analysis, and generation. Our approach combines a first-principles approach to modeling router-level connectivity with a more pragmatic use of statistics and graph theory, and we show that an improved understanding is available from viewing the Internet's router-level structure as the result of an (albeit heuristic) design process that reconciles the need for high-performance with the practical constraints imposed by the underlying technology and economics. We present empirical evidence in support of the proposed approach and its consistency with networking reality, while providing insight into the origin of high variability of inferred router-level maps. This poster reflects joint work with Lun Li, Walter Willinger, and John Doyle.

Bio: David Alderson is a postdoctoral scholar in the Division of Engineering and Applied Sciences at the California Institute of Technology (Caltech). He received a B.S.E. in Civil Engineering and Operations Research from Princeton University and the M.S. and Ph.D. degrees from the Department of Management Science and Engineering at Stanford University. His research interests include the modeling and analysis of infrastructure networks, particularly the Internet.

Abstract: The past few years have seen a tremendous interest in the phenomenon of graph clustering, a.k.a., community structure, in real world networks. These large scale structures reveal themselves as groups of vertices in which their local connections are more plentiful than their non-local ones, and their existence has implications for diverse domains such as modules in the genetic regulatory network, functional groupings of proteins, social structures and in epistemology. Most techniques, however, require that the network be relatively small and fully known. Here, we present two recently developed algorithms that handle extremely large graphs (n ~ 1,000,000) and use them to extract meaningful structural information at all scales from the recommender network of a large online retailer.

Bio: Aaron Clauset received his bachelors in Physics and Computer Science from Haverford College, and is nearing completion of his doctorate in Computer Science at the University of New Mexico. His research focuses on the inference of structural and dynamic properties of real world networks (technological and social), computational modeling, and algorithm design. In his free time, he enjoys traveling, hiking, rock climbing and photography

Abstract: Anomalies in networks such as the Internet are manifested by unexplained changes in traffic statistics over time. Reliable detection of these anomalies is hindered by the often time-varying baseline of normal traffic patterns and the need to search the high dimensional "data cube" (e.g. indexed over multi-router flow statistics sorted by source ports, destination ports, or source IP addresses) for the "needle in the haystack." Several methods have been proposed for multi-site network anomaly detection including principal components analysis (PCA) which implicitly assumes a linear model for way that anomalies enter the data. We take a different approach to this problem. The central hypothesis behind our approach is that anomalous traffic can be detected by looking for temporal changes in the local intrinsic dimension of the data space. We have developed a new model-free method for estimating this local intrinsic dimension using local k-nearest neighbor graphs (KNNG). The KNNG method can be used to segment the data over time based on the local intrinsic dimension, a method that we call complexity segmentation. In this poster we will show how complexity segmentation is applied to discovering anomalies in multi-site traffic recorded by Netflow on the Abilene backbone network.

Bio: Jose Costa received the M.S. degree in Electrical Engineering in 2002, the M.A. degree in Statistics in 2003 and the Ph.D. degree in Electrical Engineering in 2005, all from the University of Michigan, Ann Arbor. He is currently a post-doctoral fellow in the Center for the Mathematics of Information at the California Institute of Technology. His research interests include statistical signal and image processing, pattern recognition, machine learning and computational biology.

Abstract: Graphs form the foundation of many real-world datasets ranging from Internet connectivity to social networks. Yet despite this underlying structure, the size of these datasets presents a nearly insurmountable obstacle to understanding the essential character of the data. We want to understand ``what the graph looks like;'' we want to know which vertices and edges are important and what are the significant features in the graph. For a communication network, such an understanding entails recognizing the overall design of the network (e.g., hub-and-spoke, mesh, backbone), as well as identifying the “important” nodes and links. We define several compression schemes, including vertex similarity measures and vertex ranking. We present a system for condensing large graphs using both auxiliary information (such as geographic location and link type in the case of communication networks), as well as purely topological information. We examine the properties of these compression schemes, demonstrate their effects on visualization, and explore what structural graph properties they preserve when applied to both synthetic and real-world networks.

Abstract: Recent attention on the large-scale topological structure of complex networks has been heavily focused on the connectivity of network components. A particular feature of network connectivity that has generated considerable discussion is the prevalence of heavy-tails and high variability, and perhaps power laws, in node degree (e.g., number of connections). However, recent research has shown that power laws by themselves do not adequately characterize network structure. There exist many different graphs having the same node degree, some of which may be considered complete opposites from the viewpoint of particular applications. Furthermore, there are a variety of distinctly different random graph models that might give rise to a given degree distribution, and some of these models may have no domain-intrinsic meaning whatsoever. We introduce a structural metric that allows us to differentiate between all simple, connected graphs having an identical degree sequence, particularly when that sequence follows a power law. This structural view can be related to previously studied graph properties such as the various notions of self-similarity, likelihood, betweeness and assortativity. While the proposed structural metric is not intended as a general measure of all graphs, we demonstrate that it yields considerable insight into the claimed properties of so called 'scale-free' graphs. Such a view has the benefit of being minimal, in the sense that it relies on few starting assumptions, yet yields a rich and general description of the features of these networks. This poster reflects joint work with David Alderson, Walter Willinger, and John Doyle.

Bio: Lun Li received the B.S. degrees in optics and automatic control from Tsinghua University, Beijing, China, in 1999 and the M.S. degree in mechanical engineering from the University of California, Berkeley, in 2001. She has been pursuing the Ph.D. degree in electrical engineering at the California Institute of Technology, Pasadena, since 2001. Her research interests include network congestion control, routing, and Internet topology.

Abstract: Abstract: In this poster presentation, we will overview what supernetworks are, why they are needed, and the various applications including knowledge supernetworks, integrated financial and social networks, electric power supply chain networks, and supply chain networks with electronic commerce, among others, that we have been researching at the Virtual Center for Supernetworks; http://supernet.som.umass.edu We will also highlight the Braess Paradox (the original paper which has been translated from German by Braess, Nagurney, and Wakolbinger and is in press) and its relevance to transportation networks and the Internet.

Support for this research has been provided by the National Science Foundation, the AT&T Foundation, the Radcliffe Institute for Advanced Study at Harvard University, the Rockefeller Foundation, the John F. Smith Memorial Fund at the University of Massachusetts at Amherst, with additional support from the Fulbright Program.

Bio: Anna Nagurney is the John F. Smith Memorial Professor in the Department of Finance and Operations Management in the Isenberg School of Management at the University of Massachusetts at Amherst. She is also the Founding Director of the Virtual Center for Supernetworks and the Supernetworks Laboratory for Computation and Visualization at UMass Amherst. She received her AB, ScB, ScM, and PhD degrees from Brown University in Providence, Rhode Island. She devotes her career to education and research that combines management, economics, and engineering. Her focus is the applied and theoretical aspects of network systems, particularly in the areas of transportation and logistics and economics and finance. She is the editor of the book, Innovations in Financial and Economic Networks (November 2003), and has authored or co-authored 8 other books including Supernetworks: Decision-Making for the Information Age, Financial Networks, Sustainable Transportation Networks, and Network Economics, and more than 100 refereed journal articles.

Abstract: Dynamic graphs arise from massive transactional networks, such as telecommunications networks, web connectivity, credit card transactions, or online auctions. Each transaction can be represented by an edge in the graph, with nodes representing the transactors. Each edge has a time stamp associated with it, so the graphs are dynamic in the sense that they have new edges appearing regularly while old edges are getting stale. We have developed a general parameterized model for use with such graphs, and applied the model to our massive call graph at AT&T (~ 350M nodes). We also briefly show the application of the model to three projects at AT&T, finding repetitive debtors, viral marketing, and connection subgraphs.

Bio: Chris Volinsky is Director, Statistics Research Department at AT&T Research in Florham Park, NJ. Chris got his PhD from the University of Washington in 1997 (advisor: Adrian Raftery), studying Bayesian Model Averaging and its applications to survival analysis. Chris joined AT&T Labs-Research in 1997 and became Director of the Statistics Research Department in 2004. His research at AT&T focuses on analysis of massive graphs, including models for graph matching; statistical computation and visualization; and fraud detection in telecommunications. He is an avid baseball fan, enjoys traveling, and is strangely addicted to watching poker on TV.

Abstract: A recent proliferation of models of `network mechanisms' (e.g., duplication-mutation, preferential attachment, small-world models) have been proposed as models of biological networks. A recently-developed machine learning approach predicts the mechanism most accurately capturing a given network topology. We classify several biological networks, including transcriptional regulatory networks, with most attention paid to the protein-protein interaction network of Drosophila melanogaster. The machine learning approach also facilitates an information-theoretic quantitative measure of network modularity and inferring modules.