Catálogo de publicaciones - libros

We propose universal clustering in line with the concepts of universal estimation. In order to illustrate the model of universal clustering we consider family of power loss functions in probabilistic space which is marginally linked to the Kullback-Leibler divergence. The model proved to be effective in application to the synthetic data. Also, we consider large web-traffic dataset. The aim of the experiment is to explain and understand the way people interact with web sites.

A circular Self-Organising Map (SOM) based on a temporal metric has been proposed for clustering and characterising gene expressions. Expression profiles are first modelled with Radial Basis Functions. The co-expression coefficient, defined as the uncentred correlation of the differentiation of the models, is combined in a circular SOM for grouping and ordering the modelled expressions based on their temporal properties. In the proposed method the topology has been extended to temporal and cyclic ordering of the expressions. An example and a test on a microarray dataset are presented to demonstrate the advantages of the proposed method.

Recently, there has been a considerable research activity in extending topographic maps of vectorial data to more general data structures, such as sequences or trees. However, the representational capabilities and internal representations of the models are not well understood. We rigorously analyze a generalization of the Self-Organizing Map (SOM) for processing sequential data, Recursive SOM (RecSOM [1]), as a non-autonomous dynamical system consisting of a set of fixed input maps. We show that contractive fixed input maps are likely to produce Markovian organizations of receptive fields on the RecSOM map. We derive bounds on parameter (weighting the importance of importing past information when processing sequences) under which contractiveness of the fixed input maps is guaranteed.

The elastic net and related algorithms, such as generative topographic mapping, are key methods for discretized dimension-reduction problems. At their heart are priors that specify the expected topological and geometric properties of the maps. However, up to now, only a very small subset of possible priors has been considered. Here we study a much more general family originating from discrete, high-order derivative operators. We show theoretically that the form of the discrete approximation to the derivative used has a crucial influence on the resulting map. Using a new and more powerful iterative elastic net algorithm, we confirm these results empirically, and illustrate how different priors affect the form of simulated ocular dominance columns.

This paper presents a highly parallel implementation of a new type of Self-Organising Map (SOM) using graphics hardware. The Parameter-Less SOM smoothly adapts to new data while preserving the mapping formed by previous data. It is therefore in principle highly suited for interactive use, however for large data sets the computational requirements are prohibitive. This paper will present an implementation on commodity graphics hardware which uses two forms of parallelism to significantly reduce this barrier. The performance is analysed experimentally and algorithmically. An advantage to using graphics hardware is that visualisation is essentially “free”, thus increasing its suitability for interactive exploration of large data sets.

In human face recognition, different facial regions have different degrees of importance, and exploiting such information would hopefully improve the accuracy of the recognition system. A novel method is therefore proposed in this paper to automatically select the facial regions that are important for recognition. Unlike most of previous attempts, the selection is based on the facial appearance of individual subjects, rather than the appearance of all subjects. Hence the recognition process is class-specific. Experiments on the FERET face database show that the proposed methods can automatically and correctly identify those supposed important local features for recognition and thus are much beneficial to improve the recognition accuracy of the recognition system even under the condition of only one single training sample per person.

Novelty detection involves identifying novel patterns. They are not usually available during training. Even if they are, the data quantity imbalance leads to a low classification accuracy when a supervised learning scheme is employed. Thus, an unsupervised learning scheme is often employed ignoring those few novel patterns. In this paper, we propose two ways to make use of the few available novel patterns. First, a scheme to determine local thresholds for the Self Organizing Map boundary is proposed. Second, a modification of the Learning Vector Quantization learning rule is proposed so that allows one to keep codebook vectors as far from novel patterns as possible. Experimental results are quite promising.

The Self-Organising Map (SOM) is widely used to classify document collections. Such classifications are usually coarse-grained and cannot accommodate accurate document retrieval. A document classification scheme based on Multi-level Nested Self-Organising Map (MNSOM) is proposed to solve the problem. An MNSOM consists of a top map and a set of nested maps organised at different levels. The clusters on the top map of an MNSOM are at a relatively general level achieving retrieval recall, and the nested maps further elaborate the clusters into more specific groups, thus enhancing retrieval precision. The MNSOM was tested by a software document collection. The experimental results reveal that the MNSOM significantly improved the retrieval performance in comparison with the single SOM based classification.

Promiscuous human leukocyte antigen (HLA) binding peptides are ideal targets for vaccine development. Existing computational models for prediction of promiscuous peptides used hidden Markov models and artificial neural networks as prediction algorithms. We report a system based on support vector machines that outperforms previously published methods. Preliminary testing showed that it can predict peptides binding to HLA-A2 and -A3 supertype molecules with excellent accuracy, even for molecules where no binding data are currently available.

Malignancy Associated Changes are subtle changes to the nuclear texture of visually normal cells in the vicinity of a cancerous or precancerous lesion. We describe a classifier for the detection of MACs in digital images of cervical cells using artificial neural networks evolved in conjunction with an image texture feature subset. ROC curve analysis is used to compare the classification accuracy of the evolved classifier with that of standard linear discriminant analysis over the full range of classification thresholds as well as at selected optimal operating points. The nonlinear classifier does not significantly outperform the linear one, but it generalizes more readily to unseen data, and its stochastic nature provides insights into the information content of the data.