Learning from Incomplete Data

Real-world learning tasks often involve high-dimensional data sets with complex patterns of missing features. In this paper we review the problem of learning from incomplete data from two statistical perspectives---the likelihood-based and the Bayesian. The goal is two-fold: to place current neural network approaches to missing data within a statistical framework, and to describe a set of algorithms, derived from the likelihood-based framework, that handle clustering, classification, and function approximation from incomplete data in a principled and efficient manner. These algorithms are based on mixture modeling and make two distinct appeals to the Expectation-Maximization (EM) principle (Dempster, Laird, and Rubin 1977)---both for the estimation of mixture components and for coping with the missing data.

Application of compositional data analysis to geochemical data of marine sediments

In an earlier investigation (Burger et al., 2000) five sediment cores near the Rodrigues Triple Junction in the Indian Ocean were studied applying classical statistical methods (fuzzy c-means clustering, linear mixing model, principal component analysis) for the extraction of endmembers and evaluating the spatial and temporal variation of geochemical signals. Three main factors of sedimentation were expected by the marine geologists: a volcano-genetic, a hydro-hydrothermal and an ultra-basic factor. The display of fuzzy membership values and/or factor scores versus depth provided consistent results for two factors only; the ultra-basic component could not be identified. The reason for this may be that only traditional statistical methods were applied, i.e. the untransformed components were used and the cosine-theta coefficient as similarity measure. During the last decade considerable progress in compositional data analysis was made and many case studies were published using new tools for exploratory analysis of these data. Therefore it makes sense to check if the application of suitable data transformations, reduction of the D-part simplex to two or three factors and visual interpretation of the factor scores would lead to a revision of earlier results and to answers to open questions . In this paper we follow the lines of a paper of R. Tolosana- Delgado et al. (2005) starting with a problem-oriented interpretation of the biplot scattergram, extracting compositional factors, ilr-transformation of the components and visualization of the factor scores in a spatial context: The compositional factors will be plotted versus depth (time) of the core samples in order to facilitate the identification of the expected sources of the sedimentary process. Kew words: compositional data analysis, biplot, deep sea sediments

Integració a la plataforma webDTI de tècniques de clustering de fibres i visualització de clústers

Estudi, disseny i implementació de diferents tècniques d’agrupament de fibres (clustering) per tal d’integrar a la plataforma DTIWeb diferents algorismes de clustering i tècniques de visualització de clústers de fibres de forma que faciliti la interpretació de dades de DTI als especialistes

Modelo de segmentación : estudio, aplicación, ejecución y desarrollo de estrategias para la implementación de un modelo de clustering en el Sector Financiero

A finales de 2009 se emprendió un nuevo modelo de segmentación de mercados por conglomeraciones o clústers, con el cual se busca atender las necesidades de los clientes, advirtiendo el ciclo de vida en el cual se encuentran, realizando estrategias que mejoren la rentabilidad del negocio, por medio de indicadores de gestión KPI. Por medio de análisis tecnológico se desarrolló el proceso de inteligencia de la segmentación, por medio del cual se obtuvo el resultado de clústers, que poseían características similares entre sí, pero que diferían de los otros, en variables de comportamiento. Esto se refleja en el desarrollo de campañas estratégicas dirigidas que permitan crear una estrecha relación de fidelidad con el cliente, para aumentar la rentabilidad, en principio, y fortalecer la relación a largo plazo, respondiendo a la razón de ser del negocio

A new linear initialization in SOM for biomolecular data

In the past decade, the amount of data in biological field has become larger and larger; Bio-techniques for analysis of biological data have been developed and new tools have been introduced. Several computational methods are based on unsupervised neural network algorithms that are widely used for multiple purposes including clustering and visualization, i.e. the Self Organizing Maps (SOM). Unfortunately, even though this method is unsupervised, the performances in terms of quality of result and learning speed are strongly dependent from the neuron weights initialization. In this paper we present a new initialization technique based on a totally connected undirected graph, that report relations among some intersting features of data input. Result of experimental tests, where the proposed algorithm is compared to the original initialization techniques, shows that our technique assures faster learning and better performance in terms of quantization error.

Soft topographic map for clustering and classification of bacteria

In this work a new method for clustering and building a topographic representation of a bacteria taxonomy is presented. The method is based on the analysis of stable parts of the genome, the so-called “housekeeping genes”. The proposed method generates topographic maps of the bacteria taxonomy, where relations among different type strains can be visually inspected and verified. Two well known DNA alignement algorithms are applied to the genomic sequences. Topographic maps are optimized to represent the similarity among the sequences according to their evolutionary distances. The experimental analysis is carried out on 147 type strains of the Gammaprotebacteria class by means of the 16S rRNA housekeeping gene. Complete sequences of the gene have been retrieved from the NCBI public database. In the experimental tests the maps show clusters of homologous type strains and present some singular cases potentially due to incorrect classification or erroneous annotations in the database.

Bayesian clustering of farm types using the mixtures model

A Bayesian method of classifying observations that are assumed to come from a number of distinct subpopulations is outlined. The method is illustrated with simulated data and applied to the classification of farms according to their level and variability of income. The resultant classification shows a greater diversity of technical charactersitics within farm types than is conventionally the case. The range of mean farm income between groups in the new classification is wider than that of the conventional method and the variability of income within groups is narrower. Results show that the highest income group in 2000 included large specialist dairy farmers and pig and poultry producers, whilst in 2001 it included large and small specialist dairy farms and large mixed dairy and arable farms. In both years the lowest income group is dominated by non-milk producing livestock farms.

Clustering techniques applied to multiple-models structures

In this work a method for building multiple-model structures is presented. A clustering algorithm that uses data from the system is employed to define the architecture of the multiple-model, including the size of the region covered by each model, and the number of models. A heating ventilation and air conditioning system is used as a testbed of the proposed method.

Dimension and coverage of multiple-models structures using clustering techniques

In this work a method for building multiple-model structures is presented. A clustering algorithm that uses data from the system is employed to define the architecture of the multiple-model, including the size of the region covered by each model, and the number of models. A heating ventilation and air conditioning system is used as a testbed of the proposed method.

Clustering quality and topology preservation in fast learning SOMs

The Self-Organizing Map (SOM) is a popular unsupervised neural network able to provide effective clustering and data visualization for data represented in multidimensional input spaces. In this paper, we describe Fast Learning SOM (FLSOM) which adopts a learning algorithm that improves the performance of the standard SOM with respect to the convergence time in the training phase. We show that FLSOM also improves the quality of the map by providing better clustering quality and topology preservation of multidimensional input data. Several tests have been carried out on different multidimensional datasets, which demonstrate better performances of the algorithm in comparison with the original SOM.

Combined bias and outlier identification in dynamic data reconciliation

Measured process data normally contain inaccuracies because the measurements are obtained using imperfect instruments. As well as random errors one can expect systematic bias caused by miscalibrated instruments or outliers caused by process peaks such as sudden power fluctuations. Data reconciliation is the adjustment of a set of process data based on a model of the process so that the derived estimates conform to natural laws. In this paper, techniques for the detection and identification of both systematic bias and outliers in dynamic process data are presented. A novel technique for the detection and identification of systematic bias is formulated and presented. The problem of detection, identification and elimination of outliers is also treated using a modified version of a previously available clustering technique. These techniques are also combined to provide a global dynamic data reconciliation (DDR) strategy. The algorithms presented are tested in isolation and in combination using dynamic simulations of two continuous stirred tank reactors (CSTR).

Epidemic K-Means clustering

The K-Means algorithm for cluster analysis is one of the most influential and popular data mining methods. Its straightforward parallel formulation is well suited for distributed memory systems with reliable interconnection networks. However, in large-scale geographically distributed systems the straightforward parallel algorithm can be rendered useless by a single communication failure or high latency in communication paths. This work proposes a fully decentralised algorithm (Epidemic K-Means) which does not require global communication and is intrinsically fault tolerant. The proposed distributed K-Means algorithm provides a clustering solution which can approximate the solution of an ideal centralised algorithm over the aggregated data as closely as desired. A comparative performance analysis is carried out against the state of the art distributed K-Means algorithms based on sampling methods. The experimental analysis confirms that the proposed algorithm is a practical and accurate distributed K-Means implementation for networked systems of very large and extreme scale.

Ensemble clustering in deterministic ensemble Kalman filters

Ensemble clustering (EC) can arise in data assimilation with ensemble square root filters (EnSRFs) using non-linear models: an M-member ensemble splits into a single outlier and a cluster of M−1 members. The stochastic Ensemble Kalman Filter does not present this problem. Modifications to the EnSRFs by a periodic resampling of the ensemble through random rotations have been proposed to address it. We introduce a metric to quantify the presence of EC and present evidence to dispel the notion that EC leads to filter failure. Starting from a univariate model, we show that EC is not a permanent but transient phenomenon; it occurs intermittently in non-linear models. We perform a series of data assimilation experiments using a standard EnSRF and a modified EnSRF by a resampling though random rotations. The modified EnSRF thus alleviates issues associated with EC at the cost of traceability of individual ensemble trajectories and cannot use some of algorithms that enhance performance of standard EnSRF. In the non-linear regimes of low-dimensional models, the analysis root mean square error of the standard EnSRF slowly grows with ensemble size if the size is larger than the dimension of the model state. However, we do not observe this problem in a more complex model that uses an ensemble size much smaller than the dimension of the model state, along with inflation and localisation. Overall, we find that transient EC does not handicap the performance of the standard EnSRF.

Fault tolerant decentralised K-Means clustering for asynchronous large-scale networks

The K-Means algorithm for cluster analysis is one of the most influential and popular data mining methods. Its straightforward parallel formulation is well suited for distributed memory systems with reliable interconnection networks, such as massively parallel processors and clusters of workstations. However, in large-scale geographically distributed systems the straightforward parallel algorithm can be rendered useless by a single communication failure or high latency in communication paths. The lack of scalable and fault tolerant global communication and synchronisation methods in large-scale systems has hindered the adoption of the K-Means algorithm for applications in large networked systems such as wireless sensor networks, peer-to-peer systems and mobile ad hoc networks. This work proposes a fully distributed K-Means algorithm (EpidemicK-Means) which does not require global communication and is intrinsically fault tolerant. The proposed distributed K-Means algorithm provides a clustering solution which can approximate the solution of an ideal centralised algorithm over the aggregated data as closely as desired. A comparative performance analysis is carried out against the state of the art sampling methods and shows that the proposed method overcomes the limitations of the sampling-based approaches for skewed clusters distributions. The experimental analysis confirms that the proposed algorithm is very accurate and fault tolerant under unreliable network conditions (message loss and node failures) and is suitable for asynchronous networks of very large and extreme scale.

Dynamic group communication for large-scale parallel data mining

Exascale systems are the next frontier in high-performance computing and are expected to deliver a performance of the order of 10^18 operations per second using massive multicore processors. Very large- and extreme-scale parallel systems pose critical algorithmic challenges, especially related to concurrency, locality and the need to avoid global communication patterns. This work investigates a novel protocol for dynamic group communication that can be used to remove the global communication requirement and to reduce the communication cost in parallel formulations of iterative data mining algorithms. The protocol is used to provide a communication-efficient parallel formulation of the k-means algorithm for cluster analysis. The approach is based on a collective communication operation for dynamic groups of processes and exploits non-uniform data distributions. Non-uniform data distributions can be either found in real-world distributed applications or induced by means of multidimensional binary search trees. The analysis of the proposed dynamic group communication protocol has shown that it does not introduce significant communication overhead. The parallel clustering algorithm has also been extended to accommodate an approximation error, which allows a further reduction of the communication costs. The effectiveness of the exact and approximate methods has been tested in a parallel computing system with 64 processors and in simulations with 1024 processing elements.