Statistics & Probability Alumni Networking Day
Statistics & Probability Alumni Networking Day - The First Workshop in Honor of Distinguished Alumni
Statistics and Probability Alumni Networking (SPAN) Day
The First Workshop in Honor of Distinguished Alumni
Nov 5th, 9:30-5:00, Arts 101
The Department of Mathematics and Statistics and the College of Arts and Science are pleased to welcome back to campus three of its distinguished graduate alumni in the fields of probability and statistics:
Professor Haijun Li, Department of Mathematics and Statistics, Washington State University and Associate Director of the WSU Data Analytics Initiative
Professor Xikui Wang, Associate Dean, Faculty of Graduate Studies and Professor, Department of Statistics, University of Manitoba
Professor Yiqiang Zhao, Associate Dean (Research and Graduate Studies), Faculty of Science and Professor of School of Mathematics and Statistics, Carleton University
They are returning to celebrate the 30th anniversary of the start of their U of S graduate programs (September 1986) and to help us all celebrate the 100th anniversary of the U of S Alumni Association. As part of these celebrations, we are hosting a Statistics and Probability Alumni Networking (SPAN) Day on Saturday November 5, 2016. The distinguished visitors and their former graduate supervisors will give research talks and there will be an opportunity for current students and faculty to interact with the visitors. In particular, the day will end with a panel discussion on the theme: "How to turn your degree from a small graduate program into a big success."
Everyone is welcome to attend!
For additional details see http://math.usask.ca/statistics/seminar/
Please RVSP if you are interested in attending by Tuesday November 1st to Prof. Longhai Li (longhai@math.usask.ca).
PDF of Poster:
Title: Operator Regular Variation in Extreme Value Analysis
Speaker: Haijun Li, Professor
Department of Mathematics & Statistics, Washington State University
Abstract: Univariate extremes converge, with proper normalization, in distribution to one of three well-defined parametric distributions (the Frechet, Gumbel and Weibull distributions), but the parametric nature vanishes in the multivariate setting. In contrast to constructing various parametric multivariate extreme value distributions, we focus on uncovering scaling features for multivariate extremes that are useful for statistical analysis and prediction. In this talk, we will discuss a linear operator based normalization that can reveal powerful multivariate scaling properties of extremes, and show that the operator-analytic normalization provides a unified approach for high-dimensional extreme value analysis. The theory will be presented in the language of copulas and various examples will be presented in details.
Title: An overview of statistical design of clinical trials
Speaker: Xikui Wang, Associate Dean, Faculty of Graduate Studies, Professor,
Department of Statistics, University of Manitoba
Abstract: Clinical trials are regarded as the most reliable and efficient way to evaluate the efficacy of new medical interventions. This practice has taken a prominent role in modern clinical research. Clinical experimentation on human subjects requires a careful balancing act between the benefits of the collective and the benefits of the individual, as well as between efficacy and toxicity. Furthermore, a successful clinical trial must undergo Phases I to III before the new medical intervention is finally approved for marketing. In this talk, I summarize important statistical designs for Phases I to III clinical trials, with a particular focus on adaptive designs, which represent major advancements in clinical trial methodology. These designs help balance the ethical issues and improve efficiency without undermining the validity and integrity of the clinical research. The talk is based on joint work with many of my graduate students.
Title: Random Walks in the Quarter Plane and Their Applications
Speaker: Yiqiang Zhao, Associate Dean (Research & Graduate Studies),
Faculty of Science, Professor of School of Mathematics & Statistics,
Carleton University
Abstract: Behaviour of multi-dimensional random walks is one of the central focuses in probability, particularly in applied probability. Two-dimensional random walks are such examples. In this talk, we discuss stationary tail asymptotic properties for random walks in the quarter plane, which are random walks with reflective boundaries and find many important applications such as queueing systems. For a stable system, say a discrete-time ergodic two-dimensional (the state space in the quarter plane) Markov chain, the unique stationary distribution is one of the key performance metrics. However, except few very special cases, no closed form or explicit solutions for such systems. For this reason and also with its own importance, behaviour of tail asymptotics in the joint stationary distribution is a key topic in applied probability. We introduce a kernel method to study the behaviour and show that there is a total of four different types of tail asymptotics
Title: Solving Discrete-Time Discrete Event Systems by using Markov Chains
Speaker: Winfried Grassmann, Emeritus Professor,
Deptartment of Computer Science, University of Saskatchewan
The standard way to solve discrete event systems is through simulation. In this paper, we show how to find the equilibrium distribution of discrete event systems by non-Monte Carlo methods. To do this, we discretize time, and we formulate a discrete-time Markov chain representing the system. Like in discrete-event simulation, we use a next-event logic. Since the state space of this chain increases exponentially with the number of states, this method is restricted to small systems, such as 2 queues in sequence and multiserver queues with arbitrary times for arrival and service time distributions. Systems where one of the queues can reach infinity are also covered.
Title: Visualizing Independence and Irrelevance
Speaker: Mik Bickis, Emeritus Professor,
Department of Mathematics & Statistics, University of Saskatchewan
Abstract: Statistical inference is concerned with extracting relevant information from data. To know what is relevant, one must also understand what it means to be irrelevant. In classical probability theory and Bayesian inference, irrelevance is described in terms of stochastic independence. Other concepts of irrelevance are also germane to inference, and when using imprecise probabilities a variety of concepts of irrelevance and independence arise.
Properties of independence and irrelevance can be described axiomatically and studied abstractly as algebraic systems. Sets of such objects also have a geometrical structure, and exploring the geometry can stimulate intuition and elucidate some of the properties of these relations.
Title: Bent-Cable Models for Changepoint Data: Theory and Applications
Speaker: Shahedul Khan, Assistant Professor,
Department of Mathematics & Statistics, University of Saskatchewan
Abstract: TBA
Title: Impact of Misspeci ed Residual Correlation Structure on the Parameter
Estimates in a Shared Spatial Frailty Model
Speaker: Cindy Feng, Assistant Professor,
School of Public Health, University of Sasakatchewan
Abstract: In practice, survival data are collected over geographical regions, random effects corresponding to geographical regions in closer proximity to each other might also be similar in magnitude, due to underlying environmental characteristics. Therefore, shared spatial frailty model can be adopted to model the spatial correlation among the clusters, which are often implemented using Bayesian Markov Chain Monte Carlo method. This method comes at the price of slow mixing rates and heavy computational cost, which may render it impractical for data intensive application. However, a conventional frailty model assuming an independent and identically distributed (iid) frailty term can be easily and efficiently implemented in a standard statistical software. As such, we used simulations to assess the efficiency loss in fixed effect parameter estimates if residual spatial correlation is present but using a spatially uncorrelated iid frailty. Our simulation study indicates that the shared frailty model with iid frailty can still estimate the fixed covariates effects reasonably well even in the presence of residual spatial correlation, when the percentage of censoring is not too high and the number of clusters is not too low. Therefore, if the primary goal of is to assess the covariate effects, one may choose to use the simpler but less computationally intensive models with an iid frailty. This is not to say that the shared frailty model with iid frailty should be preferred over the spatial frailty model in all cases. Indeed, when the primary goal of inference is predicting the hazard for specific covariates group, additional care needs to be given due to the bias in the parameters associated with the baseline distribution.