AMS 280B, Fall 2010
Seminars in Statistical and Applied Mathematical Modeling
Lecture time: Monday, 4-5PM, E2 180
¯ AMS faculty presentations
Marc Mangel, Bruno Sanso, Athanasios Kottas, Eric Anderson
Monday, September 27, 2010, 4:00-5:00PM, E2-180
¯ AMS faculty presentations
Nicholas Brummell, Herbert Lee, Dejan Milutinovic, Raquel Prado
Monday, October 4, 2010, 4:00-5:00PM, E2-180
¯ AMS faculty presentations
Hongyun Wang, David Draper, Qi Gong, Abel Rodriguez
Monday, October 11, 2010, 4:00-5:00PM, E2-180
¯ Title: A Modern Bayesian Look at the Multi-Armed Bandit
Dr. Steven Scott, Google
Monday, October 18, 2010, 4:00-5:00PM, E2-180
A multi-armed bandit is a particular type of experimental design where the goal is to accumulate the largest possible reward. Rewards come from a payoff distribution with unknown parameters that are to be learned through sequential experimentation. This talk describes a heuristic for managing multi-armed bandits called randomized probability matching, which randomly allocates observations to arms according the Bayesian posterior probability that each arm is optimal. Advances in Bayesian computation have made randomized probability matching easy to apply to virtually any payoff distribution. This flexibility frees the experimenter to work with payoff distributions that correspond to certain classical experimental designs that have the potential to outperform methods, which are ``optimal'' in simpler contexts. We summarize the relationships between randomized probability matching and several related heuristics that have been used in the reinforcement learning literature.
Dr. Steven Scott received his PhD from the Harvard statistics department in 1998, at which point he served for 9 years on the faculty of the University of Southern California's Marshall School of Business. he left USC in 2007 to work at Capital One for a short while before coming to Google in 2008 to work with Hal Varian, Google's chief economist.
¯ Title: A near-term trajectory-based airport surface operation
Dr. Yoon Jung, NASA Ames Research Center
Monday, October 25, 2010, 4:00-5:00PM, E2-180
In recent years, NASA has been involved in airport surface traffic management research as part of NextGen air transportation technology development initiatives. The goal of the research is to develop new concepts and technologies to increase efficiency and capacity of airport surface traffic to enable super-density operations in a terminal airspace. The research portfolio includes foundational research on optimization of surface traffic management, surface trajectory modeling capabilities to enable 4-dimensional trajectory-based operations, operational concepts and algorithms for taxi conformance and conflict detection and resolution technologies (CD&R) for surface traffic, and human experiments to validate operational concepts and algorithms. The presentation will be focused on NASAÕs recent development of the concept of optimized airport taxi operations and a research prototype tool called the Spot And Runway Departure Advisor (SARDA), which is a tower controller advisory tool to provide optimal sequence and schedule advisories for departure taxiing aircraft. The recent human-in-the-loop experiment showed very promising results in terms of taxi delay reduction as well as reduction in fuel consumption and engine emissions in active movement area.
Yoon Jung is an Aerospace Engineer in Aerospace High Density Operations Branch of Aviation Systems Division at NASA Ames Research Center. Yoon has been involved in various terminal airspace ATM research since 1998 after he joined NASA. Since 2003, Yoon has been leading surface ATM research projects at NASA Ames Research Center and currently serves as an Associate Principal Investigator of the NextGen Concept and Technology development Project under Airspace Systems Program. Prior to NASA, Yoon worked as a senior engineer at Advanced Rotorcraft Technology, Mountain View, California. Yoon earned his Ph.D. in mechanical engineering from University of California, Davis, and both B.S. and M.S. degrees from Seoul National University.
¯ Title: Putting the pieces together: Patchy abundance data for long-term monitoring of (penguin) populations
Dr. Heather Lynch,
Monday, November 1st, 2010, 4:00-5:00PM, E2-180
Long-term regional scale monitoring of animal populations must always balance spatial and temporal coverage, particularly in the Antarctic where a short breeding season and challenging logistical environment limit our ability to effectively monitor the ecosystem. The Antarctic Site Inventory (ASI) is an opportunistic vessel-based survey program of the Antarctic PeninsulaÕs breeding birds. Since 1994, the ASI has completed nearly 1000 surveys at almost 130 locations spread over 180,000 km2. However, the spatiotemporal complexity of the data warrants considerable care in statistical analysis and interpretation. In this talk, I will introduce several strategies for analyzing this patchy abundance data and demonstrate how such non-standard data have facilitated important inference regarding the impact of climate change on the AntarcticÕs breeding birds.
Heather Lynch is Assistant Research Scientist in the Biology Department at the University of Maryland and Senior Research Fellow at Oceanites, Inc., a non-profit research and education foundation dedicated to monitoring and conserving the flora and fauna of the Antarctic Peninsula. Her research in quantitative ecology and environmental science is highly interdisciplinary and sits at the intersection of several specialties, including spatial statistics, mathematical biology, geography and GIS/remote sensing, and conservation biology. Though diverse, her research interests are united by a desire to apply quantitative data analysis and modeling approaches to applied problems in ecology and conservation of direct relevance to a broader resource-management or conservation audience. She is particularly interested in spatiotemporal model-data synthesis for efficient regional-scale monitoring of biodiversity.
Heather Lynch has an A.B. in Physics from Princeton University, and a M.A. in Physics and a Ph.D. in Organismic and Evolutionary Biology from Harvard University.
¯ Title: Spacecraft Dynamics, Guidance and Control: analysis, simulations, experimentation
Marcello Romano, Naval Postgraduate School
Monday, November 8, 2010, 4:00-5:00PM, E2-180
The talk will introduce some of the recent and ongoing research activities of Dr. Romano and his team. Numerical-experimental studies will be introduced, regarding the Guidance/control of spacecraft proximity maneuvers. In particular, simulations and experimental results will be presented exploiting the use of floating autonomous robots to replicate the sensors/actuator dynamics of maneuvering spacecraft in a laboratory environment. Furthermore, analytical-numerical studies will be presented, which are related to the following two subjects: 1) the use of differential drag as a propellant-less method to perform the orbital control of spacecraft in Low Earth Orbit; 2) the search for new exact solutions of the rotational motion of a rigid body.
Dr. Marcello Romano is a Tenured Associate Professor at the Naval Postgraduate School in the Mechanical and Aerospace Engineering Department, and a member of the Space Systems Academic Group. He holds a PhD (2001) and Laurea (Engineering Degree) (1997) in Aerospace Engineering from Politecnico di Milano, Italy. Before joining the faculty of Naval Postgraduate School in 2004, he has been research fellow of the US National Research Council, and visiting research associate at CERN of Geneva, Switzerland, ESA of Cologne, Germany, and Scuola Superiore SantÕAnna of Pisa, Italy. His main research interests are on the dynamics, guidance and control of spacecraft and robots. He has been conducting experimental activities with a number of advanced test beds, in laboratories, parabolic flights and on the International Space Station. He authored/co-authored 25 journal publications, 4 patent applications and a book chapter. In 2010 he received the NPS-GSEAS award for exceptional merit in research; in 2008 he was finalist faculty (top 5%) in the Scheffelin annual studentsÕ poll award at NPS; in 2006 he received the NPS Menneken annual award for excellence in scientific research. He was nominated Associate Fellow in the AIAA in 2009. He is a Senior Member in the IEEE. Dr. Romano is a member of the AIAA Technical Committee in Guidance, Navigation and Control, the AIAA Technical Committee in Space Automation and Robotics, and the IEEE Technical Committee in Space Robotics.
¯ Title: Flexible Covariance Estimation in Gaussian Graphical Models
Bala Rajaratnam, Stanford University
Monday, November 15, 2010, 4:00-5:00PM, E2-180
Covariance estimation is known to be a challenging problem, especially for high-dimensional data. In this context, graphical models can act as a tool for regularization and have proven to be excellent tools for the analysis of high dimensional data. Graphical models are statistical models where dependencies between variables are represented by means of a graph. Both frequentist and Bayesian inferential procedures for graphical models have recently received much attention in the statistics literature. The hyper-inverse Wishart distribution is a commonly used prior for Bayesian inference on covariance matrices in Gaussian Graphical models. This prior has the distinct advantage that it is a conjugate prior for this model but it suffers from lack of flexibility in high dimensional problems due to its single shape parameter. In this talk, for posterior inference on covariance matrices in decomposable Gaussian graphical models, we use a flexible class of conjugate prior distributions defined on the cone of positive-definite matrices with fixed zeros according to a graph G. This class includes the hyper inverse Wishart distribution and allows for up to k+1 shape parameters where k denotes the number of cliques in the graph. We first add to this class of priors, a reference prior, which can be viewed as an improper member of this class. We then derive the general form of the Bayes estimators under traditional loss functions adapted to graphical models and exploit the conjugacy relationship in these models to express these estimators in closed form. The closed form solutions allow us to avoid heavy computational costs that are usually incurred in these high-dimensional problems. We also investigate decision-theoretic properties of the standard frequentist estimator, which is the maximum likelihood estimator, in these problems. Furthermore, we illustrate the performance of our estimators through numerical examples and comparisons with previous work where we explore frequentist risk properties and the efficacy of graphs in the estimation of high-dimensional covariance structures. We demonstrate that our estimators yield substantial risk reductions over the maximum likelihood estimator in the graphical model.
Bala Rajaratnam is a faculty at Stanford University in the Department of Statistics. His research interests include high dimensional inference, Graphical models with applications to environmental sciences and genomics.
¯ No seminar on Monday, November 29, 2010.