In this research, we present a investigative study on how timing-based traffic analysis attacks can be used for recovery of the speech from a Voice Over Internet Protocol (VOIP) conversation by taking advantage of the reduction or suppression of the generation of traffic whenever the sender detects a voice inactivity period. We use the simple Bayesian classifier and the complex HMM (Hidden Markov Models) classier to evaluate the performance of our attack. Then we describe the usage of acoustic features in our attack to improve the performance. We conclude by presenting a number of problems that need in-depth study in order to be effective in carrying out silence detection based attacks on VOIP systems.

This thesis consists of three chapters in Bayesian financial econometrics. The first chapter proposes new dynamic component models of returns and realized covariance (RCOV) matrices based on timevarying Wishart distributions. Bayesian estimation and model comparison is conducted with a range of multivariate GARCH models and existing RCOV models from the literature. The main method of model comparison consists of a term-structure of density forecasts of returns for multiple forecast horizons. The new joint return-RCOV models provide superior density forecasts for returns from forecast horizons of 1 day to 3 months ahead as well as improved point forecasts for realized covariances. Global minimum variance portfolio selection is improved for forecast horizons up to 3 weeks out. The second chapter proposes a full Bayesian nonparametric procedure to investigate the predictive power of exchange rates on commodity prices for 3 commodity-exporting countries: Canada, Australia and New Zealand. I examine the predictive effect of exchange rates on the entire distribution of commodity prices and how this effect changes over time. A time-dependent infinite mixture of normal linear regression model is proposed for the conditional distribution of the commodity price index. The mixing weights of the mixture follow a Probit stick-breaking prior and are hence time-varying. As a result, I allow the conditional distribution of the commodity price index given exchange rates to change over time nonparametrically. The empirical study shows some new results on the predictive power of exchange rates on commodity prices. The third chapter proposes a flexible way of modeling heterogeneous breakdowns in the volatility dynamics of multivariate financial time series within the framework of MGARCH models. During periods of normal market activities, volatility dynamics are modeled by a MGARCH specification. I refer to any significant temporary deviation of the conditional covariance matrix from its implied GARCH dynamics as a covariance breakdown, which is captured through a stochastic component that allows for changes in the whole conditional covariance matrix. Bayesian inference is used and I propose an efficient posterior sampling procedure. Empirical studies show the model can capture complex and erratic temporary structural change in the volatility dynamics.

Nuclear weapons proliferation produces a vehement global safety and security concern. Perhaps most threatening is the scenario of a rogue nation or a terrorist organization acquiring nuclear weapons where the conventional ideas of nuclear deterrence may not apply. To combat this threat, innovative tools are needed that will help to improve understanding of the pathways an organization will take in attempting to obtain nuclear weapons and in predicting those pathways based on existing evidence. In this work, a methodology was developed for predicting these pathways. This methodology uses a Bayesian network. An organization’s motivations and key resources are evaluated to produce the prior probability distributions for various pathways. These probability distributions are updated as evidence is added. The methodology is implemented through the use of the commercially available Bayesian network software package, Netica. A few simple scenarios are considered to show that the model’s predictions agree with intuition. These scenarios are also used to explore the model’s strengths and limitations. The model provides a means to measure the relative threat that an organization poses to nuclear proliferation and can identify potential pathways that an organization will likely pursue. Thus, the model can serve to facilitate preventative efforts in nuclear proliferation. The model shows that an organization’s motivations biased the various pathways more than their resources; however, resources had a greater impact on an organization’s overall chance of success. Limitations of this model are that (1) it can not account for deception, (2) it can not account for parallel weapon programs, and (3) the accuracy of the output can only be as good as the user input. This work developed the first, published, quantitative methodology for predicting nuclear proliferation with consideration for how an organization’s motivations impact their pathway probabilities.

In this research, we present a investigative study on how timing-based traffic analysis attacks can be used for recovery of the speech from a Voice Over Internet Protocol (VOIP) conversation by taking advantage of the reduction or suppression of the generation of traffic whenever the sender detects a voice inactivity period. We use the simple Bayesian classifier and the complex HMM (Hidden Markov Models) classier to evaluate the performance of our attack. Then we describe the usage of acoustic features in our attack to improve the performance. We conclude by presenting a number of problems that need in-depth study in order to be effective in carrying out silence detection based attacks on VOIP systems.

Thesis (Ph. D.)--California Institute of Technology, 2003. Thesis (Ph. D.). PQ #3093487. / Includes bibliographical references. Also available via the World Wide Web. http://www.fluff.info/klemens

Thesis (Ph. D.)--University of Texas at Austin, 2002. / Vita. Includes bibliographical references. Available also from UMI Company.

I consider statistical inference for clustering, that is the arrangement of experimental units in homogeneous groups. In particular, I discuss clustering for multivariate binary outcomes. Binary data is not very informative, making it less meaningful to proceed with traditional (deterministic) clustering methods. Meaningful inference needs to account for and report the considerable uncertainty related with any reported cluster arrangement. I review and implement an approach that was proposed in the recent literature. / text

Not available / text

The main philosophy underlying this research is that a model should constitute a representation of both what we know and what we do not know about the structure and behavior of a system. In other words it should summarize, as far as possible, both our degree of certainty and degree of uncertainty, so that it facilitates statements about prediction uncertainty arising from model structural uncertainty. Based on this philosophy, the following issues were explored in the dissertation: Identification of a hydrologic system model based on assumption about perceptual and conceptual models structure only, without strong additional assumptions about its mathematical structure Development of a novel data assimilation method for extraction of mathematical relationships between modeled variables using a Bayesian probabilistic framework as an alternative to up-scaling of governing equations Evaluation of the uncertainty in predicted system response arising from three uncertainty types: o uncertainty caused by initial conditions, o uncertainty caused by inputs, o uncertainty caused by mathematical structure Merging of theory and data to identify a system as an alternative to parameter calibration and state-updating approaches Possibility of correcting existing models and including descriptions of uncertainty about their mapping relationships using the proposed method Investigation of a simple hydrological conceptual mass balance model with two-dimensional input, one-dimensional state and two-dimensional output at watershed scale and different temporal scales using the method

Genes, the fundamental building blocks of life, act together (often through their derived proteins) in modules such as protein complexes and molecular pathways to achieve a cellular function such as DNA repair and cellular transport. A current emphasis in genomics research is to identify gene modules from gene profiles, which are measurements (such as a mutant phenotype or an expression level), associated with the individual genes under conditions of interest; genes in modules often have similar gene profiles. Clustering groups of genes with similar profiles can hence deliver candidate gene modules. Pairwise similarity measures derived from these profiles are used as input to the popular hierarchical agglomerative clustering algorithms; however, these algorithms offer little guidance on how to choose candidate modules and how to improve a clustering as new data becomes available. As an alternative, there are methods based on thresholding the similarity values to obtain a graph; such a graph can be analyzed through (probabilistic) methods developed in the social sciences. However, thresholding the data discards valuable information and choosing the threshold is difficult. Extending binary relational analysis, we exploit ranked relational data as the basis for two distinct approaches for identifying modules from genomic data, both based on the theory of random graph processes. We propose probabilistic models for ranked relational data that allow candidate modules to be accompanied by objective confidence scores and that permit an elegant integration of external information on gene-gene relationships. We first followed theoretical work by Ling to objectively select exceptionally isolated groups as candidate gene modules. Secondly, inspired by stochastic block models used in the social sciences, we construct a novel model for ranked relational data, where all genes have hidden module parameters which govern the strength of all gene-gene relationships. Adapting a classical likelihood often used for the analysis of horse races, clustering is performed by estimating the module parameters using standard Bayesian methods. The method allows the incorporation of prior information on gene-gene relationships; the utility of using prior information in the form of protein-protein interaction data in clustering of yeast mutant phenotype profiles is demonstrated.