Dirichlet process mixture models for unsupervised clustering of symptoms in Parkinson's disease

In this paper, the goal of identifying disease subgroups based on differences in observed symptom profile is considered. Commonly referred to as phenotype identification, solutions to this task often involve the application of unsupervised clustering techniques. In this paper, we investigate the application of a Dirichlet Process mixture (DPM) model for this task. This model is defined by the placement of the Dirichlet Process (DP) on the unknown components of a mixture model, allowing for the expression of uncertainty about the partitioning of observed data into homogeneous subgroups. To exemplify this approach, an application to phenotype identification in Parkinson’s disease (PD) is considered, with symptom profiles collected using the Unified Parkinson’s Disease Rating Scale (UPDRS). Clustering, Dirichlet Process mixture, Parkinson’s disease, UPDRS.

Document clustering evaluation : Divergence from a random baseline

Divergence from a random baseline is a technique for the evaluation of document clustering. It ensures cluster quality measures are performing work that prevents ineffective clusterings from giving high scores to clusterings that provide no useful result. These concepts are defined and analysed using intrinsic and extrinsic approaches to the evaluation of document cluster quality. This includes the classical clusters to categories approach and a novel approach that uses ad hoc information retrieval. The divergence from a random baseline approach is able to differentiate ineffective clusterings encountered in the INEX XML Mining track. It also appears to perform a normalisation similar to the Normalised Mutual Information (NMI) measure but it can be applied to any measure of cluster quality. When it is applied to the intrinsic measure of distortion as measured by RMSE, subtraction from a random baseline provides a clear optimum that is not apparent otherwise. This approach can be applied to any clustering evaluation. This paper describes its use in the context of document clustering evaluation.

A penalty-based grouping genetic algorithm for multiple composite SaaS components clustering in Cloud

Software as a Service (SaaS) in Cloud is getting more and more significant among software users and providers recently. A SaaS that is delivered as composite application has many benefits including reduced delivery costs, flexible offers of the SaaS functions and decreased subscription cost for users. However, this approach has introduced a new problem in managing the resources allocated to the composite SaaS. The resource allocation that has been done at the initial stage may be overloaded or wasted due to the dynamic environment of a Cloud. A typical data center resource management usually triggers a placement reconfiguration for the SaaS in order to maintain its performance as well as to minimize the resource used. Existing approaches for this problem often ignore the underlying dependencies between SaaS components. In addition, the reconfiguration also has to comply with SaaS constraints in terms of its resource requirements, placement requirement as well as its SLA. To tackle the problem, this paper proposes a penalty-based Grouping Genetic Algorithm for multiple composite SaaS components clustering in Cloud. The main objective is to minimize the resource used by the SaaS by clustering its component without violating any constraint. Experimental results demonstrate the feasibility and the scalability of the proposed algorithm.

Semi-parametric modelling of ultrafine particle number concentration

Quantifying spatial and/or temporal trends in environmental modelling data requires that measurements be taken at multiple sites. The number of sites and duration of measurement at each site must be balanced against costs of equipment and availability of trained staff. The split panel design comprises short measurement campaigns at multiple locations and continuous monitoring at reference sites [2]. Here we present a modelling approach for a spatio-temporal model of ultrafine particle number concentration (PNC) recorded according to a split panel design. The model describes the temporal trends and background levels at each site. The data were measured as part of the “Ultrafine Particles from Transport Emissions and Child Health” (UPTECH) project which aims to link air quality measurements, child health outcomes and a questionnaire on the child’s history and demographics. The UPTECH project involves measuring aerosol and particle counts and local meteorology at each of 25 primary schools for two weeks and at three long term monitoring stations, and health outcomes for a cohort of students at each school [3].

Spatial and temporal analysis of Barmah Forest virus disease in Queensland, Australia

Barmah Forest virus (BFV) disease is one of the most widespread mosquito-borne diseases in Australia. The number of outbreaks and the incidence rate of BFV in Australia have attracted growing concerns about the spatio-temporal complexity and underlying risk factors of BFV disease. A large number of notifications has been recorded continuously in Queensland since 1992. Yet, little is known about the spatial and temporal characteristics of the disease. I aim to use notification data to better understand the effects of climatic, demographic, socio-economic and ecological risk factors on the spatial epidemiology of BFV disease transmission, develop predictive risk models and forecast future disease risks under climate change scenarios. Computerised data files of daily notifications of BFV disease and climatic variables in Queensland during 1992-2008 were obtained from Queensland Health and Australian Bureau of Meteorology, respectively. Projections on climate data for years 2025, 2050 and 2100 were obtained from Council of Scientific Industrial Research Organisation. Data on socio-economic, demographic and ecological factors were also obtained from relevant government departments as follows: 1) socio-economic and demographic data from Australian Bureau of Statistics; 2) wetlands data from Department of Environment and Resource Management and 3) tidal readings from Queensland Department of Transport and Main roads. Disease notifications were geocoded and spatial and temporal patterns of disease were investigated using geostatistics. Visualisation of BFV disease incidence rates through mapping reveals the presence of substantial spatio-temporal variation at statistical local areas (SLA) over time. Results reveal high incidence rates of BFV disease along coastal areas compared to the whole area of Queensland. A Mantel-Haenszel Chi-square analysis for trend reveals a statistically significant relationship between BFV disease incidence rates and age groups (ƒÓ2 = 7587, p<0.01). Semi-variogram analysis and smoothed maps created from interpolation techniques indicate that the pattern of spatial autocorrelation was not homogeneous across the state. A cluster analysis was used to detect the hot spots/clusters of BFV disease at a SLA level. Most likely spatial and space-time clusters are detected at the same locations across coastal Queensland (p<0.05). The study demonstrates heterogeneity of disease risk at a SLA level and reveals the spatial and temporal clustering of BFV disease in Queensland. Discriminant analysis was employed to establish a link between wetland classes, climate zones and BFV disease. This is because the importance of wetlands in the transmission of BFV disease remains unclear. The multivariable discriminant modelling analyses demonstrate that wetland types of saline 1, riverine and saline tidal influence were the most significant risk factors for BFV disease in all climate and buffer zones, while lacustrine, palustrine, estuarine and saline 2 and saline 3 wetlands were less important. The model accuracies were 76%, 98% and 100% for BFV risk in subtropical, tropical and temperate climate zones, respectively. This study demonstrates that BFV disease risk varied with wetland class and climate zone. The study suggests that wetlands may act as potential breeding habitats for BFV vectors. Multivariable spatial regression models were applied to assess the impact of spatial climatic, socio-economic and tidal factors on the BFV disease in Queensland. Spatial regression models were developed to account for spatial effects. Spatial regression models generated superior estimates over a traditional regression model. In the spatial regression models, BFV disease incidence shows an inverse relationship with minimum temperature, low tide and distance to coast, and positive relationship with rainfall in coastal areas whereas in whole Queensland the disease shows an inverse relationship with minimum temperature and high tide and positive relationship with rainfall. This study determines the most significant spatial risk factors for BFV disease across Queensland. Empirical models were developed to forecast the future risk of BFV disease outbreaks in coastal Queensland using existing climatic, socio-economic and tidal conditions under climate change scenarios. Logistic regression models were developed using BFV disease outbreak data for the existing period (2000-2008). The most parsimonious model had high sensitivity, specificity and accuracy and this model was used to estimate and forecast BFV disease outbreaks for years 2025, 2050 and 2100 under climate change scenarios for Australia. Important contributions arising from this research are that: (i) it is innovative to identify high-risk coastal areas by creating buffers based on grid-centroid and the use of fine-grained spatial units, i.e., mesh blocks; (ii) a spatial regression method was used to account for spatial dependence and heterogeneity of data in the study area; (iii) it determined a range of potential spatial risk factors for BFV disease; and (iv) it predicted the future risk of BFV disease outbreaks under climate change scenarios in Queensland, Australia. In conclusion, the thesis demonstrates that the distribution of BFV disease exhibits a distinct spatial and temporal variation. Such variation is influenced by a range of spatial risk factors including climatic, demographic, socio-economic, ecological and tidal variables. The thesis demonstrates that spatial regression method can be applied to better understand the transmission dynamics of BFV disease and its risk factors. The research findings show that disease notification data can be integrated with multi-factorial risk factor data to develop build-up models and forecast future potential disease risks under climate change scenarios. This thesis may have implications in BFV disease control and prevention programs in Queensland.

Eigenvoice modeling for cross likelihood ratio based speaker clustering : a Bayesian approach

This paper proposes the use of Bayesian approaches with the cross likelihood ratio (CLR) as a criterion for speaker clustering within a speaker diarization system, using eigenvoice modeling techniques. The CLR has previously been shown to be an effective decision criterion for speaker clustering using Gaussian mixture models. Recently, eigenvoice modeling has become an increasingly popular technique, due to its ability to adequately represent a speaker based on sparse training data, as well as to provide an improved capture of differences in speaker characteristics. The integration of eigenvoice modeling into the CLR framework to capitalize on the advantage of both techniques has also been shown to be beneficial for the speaker clustering task. Building on that success, this paper proposes the use of Bayesian methods to compute the conditional probabilities in computing the CLR, thus effectively combining the eigenvoice-CLR framework with the advantages of a Bayesian approach to the diarization problem. Results obtained on the 2002 Rich Transcription (RT-02) Evaluation dataset show an improved clustering performance, resulting in a 33.5% relative improvement in the overall Diarization Error Rate (DER) compared to the baseline system.

Anomalous event detection using a semi-two dimensional hidden Markov model

The rapid increase in the deployment of CCTV systems has led to a greater demand for algorithms that are able to process incoming video feeds. These algorithms are designed to extract information of interest for human operators. During the past several years, there has been a large effort to detect abnormal activities through computer vision techniques. Typically, the problem is formulated as a novelty detection task where the system is trained on normal data and is required to detect events which do not fit the learned `normal' model. Many researchers have tried various sets of features to train different learning models to detect abnormal behaviour in video footage. In this work we propose using a Semi-2D Hidden Markov Model (HMM) to model the normal activities of people. The outliers of the model with insufficient likelihood are identified as abnormal activities. Our Semi-2D HMM is designed to model both the temporal and spatial causalities of the crowd behaviour by assuming the current state of the Hidden Markov Model depends not only on the previous state in the temporal direction, but also on the previous states of the adjacent spatial locations. Two different HMMs are trained to model both the vertical and horizontal spatial causal information. Location features, flow features and optical flow textures are used as the features for the model. The proposed approach is evaluated using the publicly available UCSD datasets and we demonstrate improved performance compared to other state of the art methods.

Application of clustering technique to generating capacity state model for reliability evaluation

Capacity probability models of generating units are commonly used in many power system reliability studies, at hierarchical level one (HLI). Analytical modelling of a generating system with many units or generating units with many derated states in a system, can result in an extensive number of states in the capacity model. Limitations on available memory and computational time of present computer facilities can pose difficulties for assessment of such systems in many studies. A cluster procedure using the nearest centroid sorting method was used for IEEE-RTS load model. The application proved to be very effective in producing a highly similar model with substantially fewer states. This paper presents an extended application of the clustering method to include capacity probability representation. A series of sensitivity studies are illustrated using IEEE-RTS generating system and load models. The loss of load and energy expectations (LOLE, LOEE), are used as indicators to evaluate the application

Speaker linking using complete-linkage clustering

Speaker diarization determines instances of the same speaker within a recording. Extending this task to a collection of recordings for linking together segments spoken by a unique speaker requires speaker linking. In this paper we propose a speaker linking system using linkage clustering and state-of-the-art speaker recognition techniques. We evaluate our approach against two baseline linking systems using agglomerative cluster merging (AC) and agglomerative clustering with model retraining (ACR). We demonstrate that our linking method, using complete-linkage clustering, provides a relative improvement of 20% and 29% in attribution error rate (AER), over the AC and ACR systems, respectively.

Speaker attribution of multiple telephone conversations using a complete-linkage clustering approach

In this paper we propose and evaluate a speaker attribution system using a complete-linkage clustering method. Speaker attribution refers to the annotation of a collection of spoken audio based on speaker identities. This can be achieved using diarization and speaker linking. The main challenge associated with attribution is achieving computational efficiency when dealing with large audio archives. Traditional agglomerative clustering methods with model merging and retraining are not feasible for this purpose. This has motivated the use of linkage clustering methods without retraining. We first propose a diarization system using complete-linkage clustering and show that it outperforms traditional agglomerative and single-linkage clustering based diarization systems with a relative improvement of 40% and 68%, respectively. We then propose a complete-linkage speaker linking system to achieve attribution and demonstrate a 26% relative improvement in attribution error rate (AER) over the single-linkage speaker linking approach.

Grouping people in social networks using a weighted multi-constraints clustering method

Grouping users in social networks is an important process that improves matching and recommendation activities in social networks. The data mining methods of clustering can be used in grouping the users in social networks. However, the existing general purpose clustering algorithms perform poorly on the social network data due to the special nature of users' data in social networks. One main reason is the constraints that need to be considered in grouping users in social networks. Another reason is the need of capturing large amount of information about users which imposes computational complexity to an algorithm. In this paper, we propose a scalable and effective constraint-based clustering algorithm based on a global similarity measure that takes into consideration the users' constraints and their importance in social networks. Each constraint's importance is calculated based on the occurrence of this constraint in the dataset. Performance of the algorithm is demonstrated on a dataset obtained from an online dating website using internal and external evaluation measures. Results show that the proposed algorithm is able to increases the accuracy of matching users in social networks by 10% in comparison to other algorithms.

Image Region Salience for Improving Appearance-Based Place Recognition using a Supervised Classifier System

Many state of the art vision-based Simultaneous Localisation And Mapping (SLAM) and place recognition systems compute the salience of visual features in their environment. As computing salience can be problematic in radically changing environments new low resolution feature-less systems have been introduced, such as SeqSLAM, all of which consider the whole image. In this paper, we implement a supervised classifier system (UCS) to learn the salience of image regions for place recognition by feature-less systems. SeqSLAM only slightly benefits from the results of training, on the challenging real world Eynsham dataset, as it already appears to filter less useful regions of a panoramic image. However, when recognition is limited to specific image regions performance improves by more than an order of magnitude by utilising the learnt image region saliency. We then investigate whether the region salience generated from the Eynsham dataset generalizes to another car-based dataset using a perspective camera. The results suggest the general applicability of an image region salience mask for optimizing route-based navigation applications.

Experimental and modelling investigation of monolayer development with clustering

Standard differential equation–based models of collective cell behaviour, such as the logistic growth model, invoke a mean–field assumption which is equivalent to assuming that individuals within the population interact with each other in proportion to the average population density. Implementing such assumptions implies that the dynamics of the system are unaffected by spatial structure, such as the formation of patches or clusters within the population. Recent theoretical developments have introduced a class of models, known as moment dynamics models, which aim to account for the dynamics of individuals, pairs of individuals, triplets of individuals and so on. Such models enable us to describe the dynamics of populations with clustering, however, little progress has been made with regard to applying moment dynamics models to experimental data. Here, we report new experimental results describing the formation of a monolayer of cells using two different cell types: 3T3 fibroblast cells and MDA MB 231 breast cancer cells. Our analysis indicates that the 3T3 fibroblast cells are relatively motile and we observe that the 3T3 fibroblast monolayer forms without clustering. Alternatively, the MDA MB 231 cells are less motile and we observe that the MDA MB 231 monolayer formation is associated with significant clustering. We calibrate a moment dynamics model and a standard mean–field model to both data sets. Our results indicate that the mean–field and moment dynamics models provide similar descriptions of the 3T3 fibroblast monolayer formation whereas these two models give very different predictions for the MDA MD 231 monolayer formation. These outcomes indicate that standard mean–field models of collective cell behaviour are not always appropriate and that care ought to be exercised when implementing such a model.

The heterogeneous cluster ensemble method using hubness for clustering text documents

We propose a cluster ensemble method to map the corpus documents into the semantic space embedded in Wikipedia and group them using multiple types of feature space. A heterogeneous cluster ensemble is constructed with multiple types of relations i.e. document-term, document-concept and document-category. A final clustering solution is obtained by exploiting associations between document pairs and hubness of the documents. Empirical analysis with various real data sets reveals that the proposed meth-od outperforms state-of-the-art text clustering approaches.