2,979 publications from this institution
Feature selection is a useful preprocessing strategy when dealing with the classification and interpretation of high-dimensional biomedical data coupled with small sample sizes. A classification technique, exploiting parallelization efficiencies, is presented where sets of linear discriminant functions are designed using randomly selected feature subsets with varying cardinality. This technique, tested with biomedical data acquired from several sources, had fewer classification errors than a conventional linear discriminant analysis strategy. The algorithm development framework used to implement the technique is also discussed.
In this study, we discuss an important synergy emerging between the geometric representation of data and fuzzy logic in pattern classification. Next we show how this synergy is made operational and translates into a coherent architecture of a classifier. The crux of the proposed topology lies in a collection of simple linear classifiers (perceptrons) being combined into a logically coherent topology. In a nutshell: perceptrons come with simple geometrical interpretation while processing based on fuzzy operators (AND and OR logic units-fuzzy neurons) results in highly transparent and interpretable results. When combined together (forming a fuzzy adaptive logic network), they give rise to the computing construct that retains the advantages (processing and interpretability) of these two paradigms of information processing. We discuss a comprehensive development environment of adaptive logic networks and show their application to several classification problems.
We introduce a scalable algorithm for clustering lines and affine subspaces of fixed dimensions via data reduction using a coreset approximating scheme. A key characteristic of the proposed algorithm is exploiting a scatter-then-aggregate scheme for data partitioning, which allows to maintain a constant coreset approximation size. This key feature enables a novel algorithm which streams as well as distributes workload. We study how the proposed algorithm scales in a practical setting and compare it with other clustering methods. Results of experiments executed on the PARAM supercomputer confirm that the time complexity of algorithm is log-linear in problem size and decreases linearly as the number of processors increase. Source code for the algorithm, the experiments, and other potential applications is provided.
A large number of prediction strategies are specific to a dynamic multiobjective optimization problem (DMOP) with only one type of the Pareto set (PS) change. However, a continuous DMOP with more than one type of the unknown PS change has been seldom investigated. We present a multimodel prediction approach (MMP) realized in the framework of evolutionary algorithms (EAs) to tackle the problem. In this paper, we first detect the type of the PS change, followed by the selection of an appropriate prediction model to provide an initial population for the subsequent evolution. To observe the influence of MMP on EAs, optimal solutions obtained by three classical dynamic multiobjective EAs with and without MMP are investigated. Furthermore, to investigate the performance of MMP, three state-of-the-art prediction strategies are compared on a large number of dynamic test instances under the same particle swarm optimizer. The experimental results demonstrate that the proposed approach outperforms its counterparts under comparison on most optimization problems.
In this study, we are concerned with information granulation realized both in supervised and unsupervised mode. Our focus is on the exploitation of the technology of hyperboxes and fuzzy sets as a fundamental conceptual vehicle of information granulation. In case of supervised learning (classification), each class is described by one or more fuzzy hyperboxes defined by their corresponding minimumand maximum vertices and the corresponding hyperbox membership function. Two types of hyperboxes are formed, namely inclusion hyperboxes that contain input patterns belonging to the same class, and exclusion hyperboxes that contain patterns belonging to two or more classes, thus representing contentious areas of the pattern space. With these two types of hyperboxes each class fuzzy set is represented as a union of inclusion hyperboxes of the same class minus a union of exclusion hyperboxes. The subtraction of sets provides for efficient representation of complex topologies of pattern classes without resorting to a large number of small hyperboxes to describe each class. The proposed fuzzy hyperbox classification is compared to the original Min-Max Neural Network and the General Fuzzy Min-Max Neural Network and the origins of the improved performance of the proposed classification are identified. When it comes to the unsupervised mode of learning, we revisit a well-known method of Fuzzy C-Means (FCM) by incorporating Tchebyschev distance using which we naturally form hyperbox-like prototypes. The design of hyperbox information granules is presented and the constructs formed in this manner are evaluated with respect to their abilities to capture the structure of data.
