[Cowystats] Last Notice - ASA Spring Meeting - a few new details
pocernic at rap.ucar.edu
Wed Apr 19 09:03:03 MDT 2006
This is the last announcement for the spring meeting. Its similar to
the last announcement, with two differences. First, complete
abstracts from Dennis Cox's ASA talk and Kathe Bjork are
Second, I thought I might mention that NCAR's cafeteria only accepts
cash - no credit cards.
As promised, the weather will be perfect Friday. For
those not familiar with the Mesa lab, there are nice trails adjacent
to the building for a lunchtime walk if you choose.
Friday Forecast: Mostly sunny, with a high around 71. Calm wind.
++ Spring Meeting - April 21st - Boulder Colorado
++ Dennis Cox at Colorado School of Mines, April 21st.
Colorado/Wyoming Chapter of the American Statistical Association
Friday April 21st - 9am - 5pm.
National Center for Atmospheric Research in Boulder, CO.
The agenda for next Friday's meeting follows. We have a broad
range of topics that highlight topics being studied by the
universities in our area. The meeting is being held at NCAR's Mesa
Lab, overlooking Boulder Colorado. If you haven't visited the lab
before, it is (I think) one of the most stunning locations anywhere.
The weather should be perfect and there should be some time after lunch
to take a walk outside. We will have refreshments before the meeting
from 8:45 - 9:30 and then beer, wine and refreshments will be served
at the end of the day. Advanced registration is not required and the
event is free. Lunch is not free, but readily available in NCAR's
cafeteria. Abstracts, information on the VisLab demonstration
and direction are listed below.
Space in the VisLab is limited, so if you plan on coming to the
demonstration, please rsvp. There will be a sign-in sheet at the
security desk for people attending the meeting.
Contact me with any questions.
8:45 - 9:30 Registration - Coffee/ Donuts
9:00 - 9:30 Tour of VisLab
9:30 - 9:40 Welcome
9:40 - 10:20 Steve Sain - Models for Multivariate Spatial Lattice Data
10:20 - 10:40 Kathe Bjork University of Colorado - Health Sciences
10 minute break
10:50 - 11:00 Anthony Hayter - Introduction and overview of statistics
at University of Denver.
11:00 - 11:15 K-12 Outreach Efforts
11:15 - 12:00 Dennis Cox - P-CAP - Probability Calibration
Lunch - noon - 1:15.
1:15 - 1:30 Chapter Elections and Maurice Davies Awards
1:30 - 2:10 Katerina Kechris - Statistical Methods in Bioinformatics
2:10 - 2:30 Yu Yang - Estimating parameters for continuous-time
10 minute break
2:40 - 3:00 William Coar - Smoothing through State-Space
Models for Stream Networks
3:00 - 3:40 Thomas Lee - Pattern Generation using Likelihood
Inference for Cellular Automata
3:40 - 4:00 Caspar Ammann- Perspectives on climate of the last millennium
4- 5pm refreshments.
**** Tour of NCAR's visualization lab*
Back by popular demand, a demonstration of NCAR's VisLab
The VisLab is a state-of-the-art scientific visualization environment,
providing an immersive environment for visualizing complex datasets in
stereo-3D and collaborating across sites via AccessGrid video
teleconferencing. From a statistical perspective, the VisLab allows
data to be illustrated with both motion and depth. before the main
meeting starts, a 30 minute demonstration will be held from 9 - 9:30.
More information on the VisLab can be found at
Directions to NCAR
(Basically, in Boulder head west on Table Mesa until the road ends ~
two miles west of Broadway)
ABSTRACTS FOR SPRING MEETING
Colorado State University
Smoothing through State-Space Models for Stream Networks
Because of the natural flow of water in a stream network,
characteristics of a downstream reach may depend on characteristics of
upstream reaches. The flow of water from reach to reach provides a
natural time-like ordering throughout the stream network. By analogy
to structural time series models, we propose a local linear trend
model for a stream network and provide its expression in state-space
form. With this model, smoothed estimates generated from a variation
of the Kalman filter and smoother are the same as those obtained
through minimization of a standard penalized least squares criterion
used for spline smoothers in discrete time. That is, the state-space
formulation allows for the definition of spline smoothers on a stream
network. Estimation of the smoothing parameter is done through
maximum likelihood estimators of the variance components for the local
linear trend model.
Kathe Bjork - UCHSC
Oligonucleotide microarrays manufactured for structural and
functional genomic analyses present numerous statistical challenges, including
selection of the expression value in the low-level analysis, computations with
massive datasets, adjustment for false positives and significant background
noise, and the presence of influential and outlying gene expressions. Numerous
statistical approaches for functional genomic analyses have been published in
the biostatistical literature and some have been implemented in software
available to researchers. The investigations in this study focus on the
outcomes of mean-based and robust statistical algorithms in differential gene
expression between two tissue types, a common type of study design for
exploring the role of genetics in diseases such as cancer and heart disease.
Data for these investigations were MAS5.0 signal value (SV) and robust
multichip average (RMA) expressions derived from publicly available Affymetrix
GeneChip .CEL files, and simulations designed to represent SV and RMA
expressions. The algorithms proceed stepwise through the two types of
expression value computation, filtering for significant genes via Student's and
biweight t-tests, singular value decomposition and median polish dimension
reduction, cluster analyses, and robust sample discrimination. Results showed
that outcomes and genes detected as differentially expressed varied according
to branches followed within the algorithms, and these different
outcomes could easily be visualized. By following the steps in the complete algorithm,
investigators can distinguish genes likely to be most informative among
different tissues, and can use those genes to determine
discriminating capability by tissue type.
Dennis Cox - Rice University, Texas
A PROBABILITY CALIBRATION ASSESSMENT DISTRIBUTION
Based on an interpretation of the randomized probability integral
transform, we propose a distribution that can be used assess the
calibration of a probability model for one dimensional numerical or
ordinal data. The Probability Calibration Assessment Distribution can be
used with purely discrete models (e.g., assessing goodness of fit for a
logistic regression model) or mixed discrete/continuous models (e.g., a
model for daily rainfall, where there is a positive probability of zero
and a continuous distribution for positive values). Several examples from
a variety of applications will be presented to indicate how the results
can be interpreted. It will be shown how to implement a goodness of fit
test in certain situations. A few results and open problems on the
asymptotics of the associated empirical process will be discussed.
Note, Dennis Cox will also be speaking in the afternoon at the School
of Mines in Golden)
Department of Preventive Medicine and Biometrics
University of Colorado at Denver and Health Sciences Center
"Statistical Methods in Bioinformatics"
In the last few decades, new experimental technology in the life sciences
has created an exponentially growing quantity of biological data. The
challenge for the statistics community is to organize this increasing
amount of knowledge to help answer questions about biological systems and
processes. In this talk, I will give an overview of statistical methods in
several areas of bioinformatics research.
Colorado State University
Pattern Generation using Likelihood Inference for Cellular Automata
Cellular automata (CA) is a dynamical system that evolves on a
discrete lattice. In this talk CA is applied to model and generate
various binary image textures. The idea is to, given an observed
binary texture image, estimate the unknown CA "rule" that generated
the image. As to be demonstrated in this talk, this estimation
problem can be posted as a statistical model selection problem, and
the minimum description length principle is adopted to provide a
University of Colorado - Denver
Models for Multivariate Spatial Lattice Data
Many spatial problems, particularly those concerning environmental
investigations, are inherently multivariate, in that more than one
variable is typically measured at each spatial location. Focusing on
spatial lattice data, this talk will cover the details of a multivariate
Markov random field model (also referred to as a conditional
autoregressive or CAR model). Attention will be given to the spatial
covariance structure, in particular to the potential asymmetry in the
spatial cross-covariances. Incoparting these multivariate Markov random
field models into a hierarchical framework, the models will be
demonstrated on examples including combining climate model output for
assessing climate change and analyzing the racial distribution of
traffic and pedestrian stops by police in the city of Denver.
Colorado State University
Estimating parameters for continuous-time autoregression models
In this talk, we consider the problem of estimating the parameters of a
continuous-time autoregression (linear and non-linear), based on a closely and
regularly spaced time series. Our method consists of two parts. We first
calculate the exact maximum likelihood estimators of the autoregressive
coefficients, conditional on the initial observations and assuming the process
is observed continuously. There is a close form expression for these
estimators. Then we approximate the exact solution using the observations
which are sampled at discrete times. This procedure can be carried out for
both linear and non-linear autoregressions. And our simulation results
demonstrate the good performance of it.
### Dennis Cox at Colorado School of Mines
Chauvenet Hall 143 3:00-4:00 pm.
Convergence of Gibbs Measures Associated with Simulated Annealing
Dr. Cox's visit to Colorado is sponsored by the School of Mines.
While this gives us the opportunity to have Dennis speak at the
chapter meeting in the morning, it creates a conflict in the
afternoon. On the bright side of things, there is no reason
not to see a interesting statistical talk on the 21st.
The motivating application for this research concerns modeling the
equilibrium properties of functional materials including shape memory
alloys. This leads one to seek solutions of differential inclusions:
find a function satisfying given boundary conditions whose derivative
is allowed to take on values from a set of allowed values
corresponding to allowable crystalline configurations of the
material. There are corresponding variational problems, but the
practical solution of these has proven notoriously difficult. One
promising approach is Simulated Annealing, a stochastic optimization
algorithm, but computational experience suggests that it has major
problems. We present here an analysis of the probability measures of
the simulated annealing algorithm that shows they can converge to an
National Center for Atmospheric Research
Research Applications Laboratory
More information about the Cowystats