|
Laboratory for Research in Statistics
and Probability
|
Location: Carleton University (Macphail Room, 4351 Herzberg Building)
Coordinator: Antal A. Járai
Note: The seminar usually meets every second week.
Past seminars: Fall 2004
Speaker:
Shoja'eddin Chenouri (Carleton)
Title:
Data depth, theory and applications; a review
Abstract:
Multivariate analysis plays a role of ever-increasing importance in
statistics. Most statistical experiments are multivariate in nature, and
large scale multivariate datasets are now made tractable by recent
explosive advances in computer technology. However, classical
multivariate analysis relies heavily on the assumption of normality or
near normality, which is often difficult to justify in practice.
The goal of this talk is to give an introduction to the recent
advancements in multivariate nonparametric methodology based on the
concept of data depth. A data depth is a measure of how deep or outlying
a given point is with respect to a data cloud or a distribution. Depth
functions introduce center-outward orderings and rankings of
multidimensional data.
In this talk, we shall review briefly different notions of data depth,
the associated multidimensional medians, and their statistical
properties and computational complexities. Along with examples we shall
introduce some graphical techniques for the multivariate goodness of
fit, multivariate dispersion, and skewness, etc. We also construct
nonparametric multivariate multi-sample location and dispersion tests.
To conclude some other potential applications in data analysis and also
research topics related to the data depth will be discussed.
Speaker:
Keivan Navaie (Postdoctoral fellow, Carleton)
Title:
Application of time series theory in performance evaluation of
next generation wireless cellular networks
Abstract:
In this presentation, we address an application of time-series theory in
modelling of the next generation wireless cellular communication systems. We
first briefly discuss the basic concepts of wireless communications. Then we
formulat the wireless cellular network performance indicator as a stochastic
process that depends on the traffic characteristics of users and wireless
channel variations. Traffic characteristics of the users include call arrival
rate, their call duration and bit rate variations.
We then show that the wireless network performance indicator in multi-service
wireless networks follows an asymptotically self-similar (as-s) process. Here
the as-s model is valid under certain conditions on channel variations and
traffic characteristics that cover a range of practical situations. We derive
these conditions and generalize earlier results, obtained for data-centric
wireless cellular networks, to multi-service wireless cellular networks.
Simulation results are then presented for actual cases that confirm analytical
results. Furthermore, we discuss the impact of the developed analysis on
designing appropriate mechanisms for controlling wireless network resources.
Speaker:
Avinash Singh (Statistics Canada and Carleton)
Title:
Protecting Quality and Confidentiality of Micro Data by MASSC: A Review
Abstract:
The reputation and credibility of a data producer are at stake if the respondents
in the database do not have confidence in the confidentiality of their sensitive
information. In an important scenario known as "disclosure by response knowledge,"
the respondent identifies his own record from the database and is concerned that
he could be detected by someone who might know enough about the respondent to
identify his record. Although this intrusion scenario may be deemed very stringent,
as the threat of intrusion may be only a perception and not real, it is nevertheless
very practical because it is the respondent's trust the producer wants to maintain
and any perceived breach of trust might lead to a lack of cooperation from the
respondent.
The MASSC (Micro Agglomeration with Substitution, Subsampling, and Calibration)
method of Singh (2002), and Singh, Yu, and Dunteman (2003) addresses the above problem
and the more general disclosure scenario of inside intrusion where the intruder knows
the presence of his target in the database as well as information about some
identifying variables. MASSC uses optimal survey sampling methods for minimizing cost
subject to bias and precision constraints in order to provide simultaneous control
on disclosure risk and information loss. The method consists of four steps
corresponding to Micro Agglomeration for partitioning the database into risk strata,
optimal probabilistic Substitution for perturbation to introduce uncertainty about
the identity of a record, optimal probabilistic Subsampling for suppression to
introduce uncertainty about the presence of a record, and optimal sampling weight
Calibration for preserving estimates of key study variables in the treated database.
The main purpose of this talk is to give a review of MASSC followed by a simple
illustrative example to explain the steps of MASSC. An application to a real data
from the National survey on drug use and health is also presented.
Speaker:
Marie-Claude Viano (Lille 1 University)
Title:
Almost periodically correlated processes with long-memory
Abstract: This lecture, from a joint paper with Anne Philippe and Donatas Surgailis, overviews some models of periodically correlated and almost periodically correlated time series presenting long-range dependence. We propose four classes of processes for which we study the behavior of the covariance and evaluate the Hurst index. We prove the convergence of the Donsker lines. I intend to present some simulations.