Probabilistic multiple model neural network based leak detection system:experimental study

This paper presents an effective decision making system for leak detection based on multiple generalized linear models and clustering techniques. The training data for the proposed decision system is obtained by setting up an experimental pipeline fully operational distribution system. The system is also equipped with data logging for three variables; namely, inlet pressure, outlet pressure, and outlet flow. The experimental setup is designed such that multi-operational conditions of the distribution system, including multi pressure and multi flow can be obtained. We then statistically tested and showed that pressure and flow variables can be used as signature of leak under the designed multi-operational conditions. It is then shown that the detection of leakages based on the training and testing of the proposed multi model decision system with pre data clustering, under multi operational conditions produces better recognition rates in comparison to the training based on the single model approach. This decision system is then equipped with the estimation of confidence limits and a method is proposed for using these confidence limits for obtaining more robust leakage recognition results.

Detection of foveation windows and analysis of foveation sequences in congenital nystagmus

Congenital nystagmus (CN) is an ocular-motor disorder characterised by involuntary, conjugated ocular oscillations, that can arise since the first months of life. Pathogenesis of congenital nystagmus is still under investigation. In general, CN patients show a considerable decrease of their visual acuity: image fixation on the retina is disturbed by nystagmus continuous oscillations, mainly horizontal. However, image stabilisation is still achieved during the short periods in which eye velocity slows down while the target image is placed onto the fovea (called foveation intervals). To quantify the extent of nystagmus, eye movement recording are routinely employed, allowing physicians to extract and analyse nystagmus main features such as shape, amplitude and frequency. Using eye movement recording, it is also possible to compute estimated visual acuity predictors: analytical functions which estimates expected visual acuity using signal features such as foveation time and foveation position variability. Use of those functions add information to typical visual acuity measurement (e.g. Landolt C test) and could be a support for therapy planning or monitoring. This study focus on robust detection of CN patients' foveations. Specifically, it proposes a method to recognize the exact signal tracts in which a subject foveates, This paper also analyses foveation sequences. About 50 eyemovement recordings, either infrared-oculographic or electrooculographic, from different CN subjects were acquired. Results suggest that an exponential interpolation for the slow phases of nystagmus could improve foveation time computing and reduce influence of breaking saccades and data noise. Moreover a concise description of foveation sequence variability can be achieved using non-fitting splines. © 2009 Springer Berlin Heidelberg.

Early Detection of Emergent Events Based on an Extremal Process Approach

2000 Mathematics Subject Classification: Primary 60G70, 62F03.

Multiresponse Robust Engineering: Case with Errors in Factor Levels

2000 Mathematics Subject Classification: 62J05, 62J10, 62F35, 62H12, 62P30.

Detection and Classification of Whale Acoustic Signals

This dissertation focuses on two vital challenges in relation to whale acoustic signals: detection and classification.

In detection, we evaluated the influence of the uncertain ocean environment on the spectrogram-based detector, and derived the likelihood ratio of the proposed Short Time Fourier Transform detector. Experimental results showed that the proposed detector outperforms detectors based on the spectrogram. The proposed detector is more sensitive to environmental changes because it includes phase information.

In classification, our focus is on finding a robust and sparse representation of whale vocalizations. Because whale vocalizations can be modeled as polynomial phase signals, we can represent the whale calls by their polynomial phase coefficients. In this dissertation, we used the Weyl transform to capture chirp rate information, and used a two dimensional feature set to represent whale vocalizations globally. Experimental results showed that our Weyl feature set outperforms chirplet coefficients and MFCC (Mel Frequency Cepstral Coefficients) when applied to our collected data.

Since whale vocalizations can be represented by polynomial phase coefficients, it is plausible that the signals lie on a manifold parameterized by these coefficients. We also studied the intrinsic structure of high dimensional whale data by exploiting its geometry. Experimental results showed that nonlinear mappings such as Laplacian Eigenmap and ISOMAP outperform linear mappings such as PCA and MDS, suggesting that the whale acoustic data is nonlinear.

We also explored deep learning algorithms on whale acoustic data. We built each layer as convolutions with either a PCA filter bank (PCANet) or a DCT filter bank (DCTNet). With the DCT filter bank, each layer has different a time-frequency scale representation, and from this, one can extract different physical information. Experimental results showed that our PCANet and DCTNet achieve high classification rate on the whale vocalization data set. The word error rate of the DCTNet feature is similar to the MFSC in speech recognition tasks, suggesting that the convolutional network is able to reveal acoustic content of speech signals.

Bayesian Inference in Large-scale Problems

Many modern applications fall into the category of "large-scale" statistical problems, in which both the number of observations n and the number of features or parameters p may be large. Many existing methods focus on point estimation, despite the continued relevance of uncertainty quantification in the sciences, where the number of parameters to estimate often exceeds the sample size, despite huge increases in the value of n typically seen in many fields. Thus, the tendency in some areas of industry to dispense with traditional statistical analysis on the basis that "n=all" is of little relevance outside of certain narrow applications. The main result of the Big Data revolution in most fields has instead been to make computation much harder without reducing the importance of uncertainty quantification. Bayesian methods excel at uncertainty quantification, but often scale poorly relative to alternatives. This conflict between the statistical advantages of Bayesian procedures and their substantial computational disadvantages is perhaps the greatest challenge facing modern Bayesian statistics, and is the primary motivation for the work presented here.

Two general strategies for scaling Bayesian inference are considered. The first is the development of methods that lend themselves to faster computation, and the second is design and characterization of computational algorithms that scale better in n or p. In the first instance, the focus is on joint inference outside of the standard problem of multivariate continuous data that has been a major focus of previous theoretical work in this area. In the second area, we pursue strategies for improving the speed of Markov chain Monte Carlo algorithms, and characterizing their performance in large-scale settings. Throughout, the focus is on rigorous theoretical evaluation combined with empirical demonstrations of performance and concordance with the theory.

One topic we consider is modeling the joint distribution of multivariate categorical data, often summarized in a contingency table. Contingency table analysis routinely relies on log-linear models, with latent structure analysis providing a common alternative. Latent structure models lead to a reduced rank tensor factorization of the probability mass function for multivariate categorical data, while log-linear models achieve dimensionality reduction through sparsity. Little is known about the relationship between these notions of dimensionality reduction in the two paradigms. In Chapter 2, we derive several results relating the support of a log-linear model to nonnegative ranks of the associated probability tensor. Motivated by these findings, we propose a new collapsed Tucker class of tensor decompositions, which bridge existing PARAFAC and Tucker decompositions, providing a more flexible framework for parsimoniously characterizing multivariate categorical data. Taking a Bayesian approach to inference, we illustrate empirical advantages of the new decompositions.

Latent class models for the joint distribution of multivariate categorical, such as the PARAFAC decomposition, data play an important role in the analysis of population structure. In this context, the number of latent classes is interpreted as the number of genetically distinct subpopulations of an organism, an important factor in the analysis of evolutionary processes and conservation status. Existing methods focus on point estimates of the number of subpopulations, and lack robust uncertainty quantification. Moreover, whether the number of latent classes in these models is even an identified parameter is an open question. In Chapter 3, we show that when the model is properly specified, the correct number of subpopulations can be recovered almost surely. We then propose an alternative method for estimating the number of latent subpopulations that provides good quantification of uncertainty, and provide a simple procedure for verifying that the proposed method is consistent for the number of subpopulations. The performance of the model in estimating the number of subpopulations and other common population structure inference problems is assessed in simulations and a real data application.

In contingency table analysis, sparse data is frequently encountered for even modest numbers of variables, resulting in non-existence of maximum likelihood estimates. A common solution is to obtain regularized estimates of the parameters of a log-linear model. Bayesian methods provide a coherent approach to regularization, but are often computationally intensive. Conjugate priors ease computational demands, but the conjugate Diaconis--Ylvisaker priors for the parameters of log-linear models do not give rise to closed form credible regions, complicating posterior inference. In Chapter 4 we derive the optimal Gaussian approximation to the posterior for log-linear models with Diaconis--Ylvisaker priors, and provide convergence rate and finite-sample bounds for the Kullback-Leibler divergence between the exact posterior and the optimal Gaussian approximation. We demonstrate empirically in simulations and a real data application that the approximation is highly accurate, even in relatively small samples. The proposed approximation provides a computationally scalable and principled approach to regularized estimation and approximate Bayesian inference for log-linear models.

Another challenging and somewhat non-standard joint modeling problem is inference on tail dependence in stochastic processes. In applications where extreme dependence is of interest, data are almost always time-indexed. Existing methods for inference and modeling in this setting often cluster extreme events or choose window sizes with the goal of preserving temporal information. In Chapter 5, we propose an alternative paradigm for inference on tail dependence in stochastic processes with arbitrary temporal dependence structure in the extremes, based on the idea that the information on strength of tail dependence and the temporal structure in this dependence are both encoded in waiting times between exceedances of high thresholds. We construct a class of time-indexed stochastic processes with tail dependence obtained by endowing the support points in de Haan's spectral representation of max-stable processes with velocities and lifetimes. We extend Smith's model to these max-stable velocity processes and obtain the distribution of waiting times between extreme events at multiple locations. Motivated by this result, a new definition of tail dependence is proposed that is a function of the distribution of waiting times between threshold exceedances, and an inferential framework is constructed for estimating the strength of extremal dependence and quantifying uncertainty in this paradigm. The method is applied to climatological, financial, and electrophysiology data.

The remainder of this thesis focuses on posterior computation by Markov chain Monte Carlo. The Markov Chain Monte Carlo method is the dominant paradigm for posterior computation in Bayesian analysis. It has long been common to control computation time by making approximations to the Markov transition kernel. Comparatively little attention has been paid to convergence and estimation error in these approximating Markov Chains. In Chapter 6, we propose a framework for assessing when to use approximations in MCMC algorithms, and how much error in the transition kernel should be tolerated to obtain optimal estimation performance with respect to a specified loss function and computational budget. The results require only ergodicity of the exact kernel and control of the kernel approximation accuracy. The theoretical framework is applied to approximations based on random subsets of data, low-rank approximations of Gaussian processes, and a novel approximating Markov chain for discrete mixture models.

Data augmentation Gibbs samplers are arguably the most popular class of algorithm for approximately sampling from the posterior distribution for the parameters of generalized linear models. The truncated Normal and Polya-Gamma data augmentation samplers are standard examples for probit and logit links, respectively. Motivated by an important problem in quantitative advertising, in Chapter 7 we consider the application of these algorithms to modeling rare events. We show that when the sample size is large but the observed number of successes is small, these data augmentation samplers mix very slowly, with a spectral gap that converges to zero at a rate at least proportional to the reciprocal of the square root of the sample size up to a log factor. In simulation studies, moderate sample sizes result in high autocorrelations and small effective sample sizes. Similar empirical results are observed for related data augmentation samplers for multinomial logit and probit models. When applied to a real quantitative advertising dataset, the data augmentation samplers mix very poorly. Conversely, Hamiltonian Monte Carlo and a type of independence chain Metropolis algorithm show good mixing on the same dataset.

Statistically robust representation and comparison of mortality profiles in archaeozoology

Archaeozoological mortality profiles have been used to infer site-specific subsistence strategies. There is however no common agreement on the best way to present these profiles and confidence intervals around age class proportions. In order to deal with these issues, we propose the use of the Dirichlet distribution and present a new approach to perform age-at-death multivariate graphical comparisons. We demonstrate the efficiency of this approach using domestic sheep/goat dental remains from 10 Cardial sites (Early Neolithic) located in South France and the Iberian Peninsula. We show that the Dirichlet distribution in age-at-death analysis can be used: (i) to generate Bayesian credible intervals around each age class of a mortality profile, even when not all age classes are observed; and (ii) to create 95% kernel density contours around each age-at-death frequency distribution when multiple sites are compared using correspondence analysis. The statistical procedure we present is applicable to the analysis of any categorical count data and particularly well-suited to archaeological data (e.g. potsherds, arrow heads) where sample sizes are typically small.

Clustering in non-parametric multivariate analyses.

Non-parametric multivariate analyses of complex ecological datasets are widely used. Following appropriate pre-treatment of the data inter-sample resemblances are calculated using appropriate measures. Ordination and clustering derived from these resemblances are used to visualise relationships among samples (or variables). Hierarchical agglomerative clustering with group-average (UPGMA) linkage is often the clustering method chosen. Using an example dataset of zooplankton densities from the Bristol Channel and Severn Estuary, UK, a range of existing and new clustering methods are applied and the results compared. Although the examples focus on analysis of samples, the methods may also be applied to species analysis. Dendrograms derived by hierarchical clustering are compared using cophenetic correlations, which are also used to determine optimum  in flexible beta clustering. A plot of cophenetic correlation against original dissimilarities reveals that a tree may be a poor representation of the full multivariate information. UNCTREE is an unconstrained binary divisive clustering algorithm in which values of the ANOSIM R statistic are used to determine (binary) splits in the data, to form a dendrogram. A form of flat clustering, k-R clustering, uses a combination of ANOSIM R and Similarity Profiles (SIMPROF) analyses to determine the optimum value of k, the number of groups into which samples should be clustered, and the sample membership of the groups. Robust outcomes from the application of such a range of differing techniques to the same resemblance matrix, as here, result in greater confidence in the validity of a clustering approach.

Clustering in non-parametric multivariate analyses.

Non-parametric multivariate analyses of complex ecological datasets are widely used. Following appropriate pre-treatment of the data inter-sample resemblances are calculated using appropriate measures. Ordination and clustering derived from these resemblances are used to visualise relationships among samples (or variables). Hierarchical agglomerative clustering with group-average (UPGMA) linkage is often the clustering method chosen. Using an example dataset of zooplankton densities from the Bristol Channel and Severn Estuary, UK, a range of existing and new clustering methods are applied and the results compared. Although the examples focus on analysis of samples, the methods may also be applied to species analysis. Dendrograms derived by hierarchical clustering are compared using cophenetic correlations, which are also used to determine optimum  in flexible beta clustering. A plot of cophenetic correlation against original dissimilarities reveals that a tree may be a poor representation of the full multivariate information. UNCTREE is an unconstrained binary divisive clustering algorithm in which values of the ANOSIM R statistic are used to determine (binary) splits in the data, to form a dendrogram. A form of flat clustering, k-R clustering, uses a combination of ANOSIM R and Similarity Profiles (SIMPROF) analyses to determine the optimum value of k, the number of groups into which samples should be clustered, and the sample membership of the groups. Robust outcomes from the application of such a range of differing techniques to the same resemblance matrix, as here, result in greater confidence in the validity of a clustering approach.

Point-and-shoot: rapid quantitative detection methods for on-site food fraud analysis – moving out of the laboratory and into the food supply chain

Major food adulteration and contamination events occur with alarming regularity and are known to be episodic, with the question being not if but when another large-scale food safety/integrity incident will occur. Indeed, the challenges of maintaining food security are now internationally recognised. The ever increasing scale and complexity of food supply networks can lead to them becoming significantly more vulnerable to fraud and contamination, and potentially dysfunctional. This can make the task of deciding which analytical methods are more suitable to collect and analyse (bio)chemical data within complex food supply chains, at targeted points of vulnerability, that much more challenging. It is evident that those working within and associated with the food industry are seeking rapid, user-friendly methods to detect food fraud and contamination, and rapid/high-throughput screening methods for the analysis of food in general. In addition to being robust and reproducible, these methods should be portable and ideally handheld and/or remote sensor devices, that can be taken to or be positioned on/at-line at points of vulnerability along complex food supply networks and require a minimum amount of background training to acquire information rich data rapidly (ergo point-and-shoot). Here we briefly discuss a range of spectrometry and spectroscopy based approaches, many of which are commercially available, as well as other methods currently under development. We discuss a future perspective of how this range of detection methods in the growing sensor portfolio, along with developments in computational and information sciences such as predictive computing and the Internet of Things, will together form systems- and technology-based approaches that significantly reduce the areas of vulnerability to food crime within food supply chains. As food fraud is a problem of systems and therefore requires systems level solutions and thinking.

Multivariate analysis of regional-scale geochemical data for environmental monitoring

A compositional multivariate approach is used to analyse regional scale soil geochemical data obtained as part of the Tellus Project generated by the Geological Survey Northern Ireland (GSNI). The multi-element total concentration data presented comprise XRF analyses of 6862 rural soil samples collected at 20cm depths on a non-aligned grid at one site per 2 km2. Censored data were imputed using published detection limits. Using these imputed values for 46 elements (including LOI), each soil sample site was assigned to the regional geology map provided by GSNI initially using the dominant lithology for the map polygon. Northern Ireland includes a diversity of geology representing a stratigraphic record from the Mesoproterozoic, up to and including the Palaeogene. However, the advance of ice sheets and their meltwaters over the last 100,000 years has left at least 80% of the bedrock covered by superficial deposits, including glacial till and post-glacial alluvium and peat. The question is to what extent the soil geochemistry reflects the underlying geology or superficial deposits. To address this, the geochemical data were transformed using centered log ratios (clr) to observe the requirements of compositional data analysis and avoid closure issues. Following this, compositional multivariate techniques including compositional Principal Component Analysis (PCA) and minimum/maximum autocorrelation factor (MAF) analysis method were used to determine the influence of underlying geology on the soil geochemistry signature. PCA showed that 72% of the variation was determined by the first four principal components (PC’s) implying “significant” structure in the data. Analysis of variance showed that only 10 PC’s were necessary to classify the soil geochemical data. To consider an improvement over PCA that uses the spatial relationships of the data, a classification based on MAF analysis was undertaken using the first 6 dominant factors. Understanding the relationship between soil geochemistry and superficial deposits is important for environmental monitoring of fragile ecosystems such as peat. To explore whether peat cover could be predicted from the classification, the lithology designation was adapted to include the presence of peat, based on GSNI superficial deposit polygons and linear discriminant analysis (LDA) undertaken. Prediction accuracy for LDA classification improved from 60.98% based on PCA using 10 principal components to 64.73% using MAF based on the 6 most dominant factors. The misclassification of peat may reflect degradation of peat covered areas since the creation of superficial deposit classification. Further work will examine the influence of underlying lithologies on elemental concentrations in peat composition and the effect of this in classification analysis.

A field experiment on a steel Gerber-truss bridge for damage detection utilizing vehicle-induced vibrations

A field experiment was conducted on a real continuous steel Gerber-truss bridge with artificial damage applied. This article summarizes the results of the experiment for bridge damage detection utilizing traffic-induced vibrations. It investigates the sensitivities of a number of quantities to bridge damage including the identified modal parameters and their statistical patterns, Nair’s damage indicator and its statistical pattern and different sets of measurement points. The modal parameters are identified by autoregressive time-series models. The decision on bridge health condition is made and the sensitivity of variables is evaluated with the aid of the Mahalanobis–Taguchi system, a multivariate pattern recognition tool. Several observations are made as follows. For the modal parameters, although bridge damage detection can be achieved by performing Mahalanobis–Taguchi system on certain modal parameters of certain sets of measurement points, difficulties were faced in subjective selection of meaningful bridge modes and low sensitivity of the statistical pattern of the modal parameters to damage. For Nair’s damage indicator, bridge damage detection could be achieved by performing Mahalanobis–Taguchi system on Nair’s damage indicators of most sets of measurement points. As a damage indicator, Nair’s damage indicator was superior to the modal parameters. Three main advantages were observed: it does not require any subjective decision in calculating Nair’s damage indicator, thus potential human errors can be prevented and an automatic detection task can be achieved; its statistical pattern has high sensitivity to damage and, finally, it is flexible regarding the choice of sets of measurement points.

A new generation of companion diagnostics: cobas BRAF, KRAS and EGFR mutation detection tests

The cobas® (Roche) portfolio of companion diagnostics in oncology currently has three assays CE-marked for in vitro diagnostics. Two of these (EGFR and BRAF) are also US FDA-approved. These assays detect clinically relevant mutations that are correlated with response (BRAF, EGFR) or lack of response (KRAS) to targeted therapies such as selective mutant BRAF inhibitors in malignant melanoma, tyrosine kinases inhibitor in non-small cell lung cancer and anti-EGFR monoclonal antibodies in colorectal cancer, respectively. All these assays are run on a single platform using DNA extracted from a single 5 µm section of a formalin-fixed paraffin-embedded tissue block. The assays provide an ‘end-to-end’ solution from extraction of DNA to automated analysis and report on the cobas z 480. The cobas tests have shown robust and reproducible performance, with high sensitivity and specificity and low limit of detection, making them suitable as companion diagnostics for clinical use.

Feature Selection for Anomaly Detection Using Optical Emission Spectroscopy

To maintain the pace of development set by Moore's law, production processes in semiconductor manufacturing are becoming more and more complex. The development of efficient and interpretable anomaly detection systems is fundamental to keeping production costs low. As the dimension of process monitoring data can become extremely high anomaly detection systems are impacted by the curse of dimensionality, hence dimensionality reduction plays an important role. Classical dimensionality reduction approaches, such as Principal Component Analysis, generally involve transformations that seek to maximize the explained variance. In datasets with several clusters of correlated variables the contributions of isolated variables to explained variance may be insignificant, with the result that they may not be included in the reduced data representation. It is then not possible to detect an anomaly if it is only reflected in such isolated variables. In this paper we present a new dimensionality reduction technique that takes account of such isolated variables and demonstrate how it can be used to build an interpretable and robust anomaly detection system for Optical Emission Spectroscopy data.

Fast and robust face finding via local context

[EN]In visual surveillance face detection can be an important cue for initializing tracking algorithms. Recent work in psychophics hints at the importance of the local context of a face for robust detection, such as head contours and torso. This paper describes a detector that actively utilizes the idea of local context. The promise is to gain robustness that goes beyond the capabilities of traditional face detection making it particularly interesting for surveillance. The performance of the proposed detector in terms of accuracy and speed is evaluated on data sets from PETS 2000 and PETS 2003 and compared to the object-centered approach. Particular attention is paid to the role of available image resolution.