Independent Component Analysis (ICA) has emerged as a useful method for separation of components, such as in removing noise from data. We examine one of the challenges of ICA - instability, particularly in high dimensions, when the independent components vary, each time when ICA is performed. This may be due to various causes including the stochastic nature of the algorithm and the additive noise. The objective of this study is to examine denoising and stability issues of ICA in high dimensions and make a comparative evaluation of select approaches. We take a challenging electrocardiogram dataset which is a high-dimensional time series of multiple sensors. We experiment with a mix of approaches and methods - for resampling, clustering, ICA algorithms and dimensionality. We check the internal validity using the Icasso stability index, the Amari separation performance index and the Minimum Distance (MD) index. The first key contribution of this work is that it finds counter-evidence to the claim that resampling (bootstrapping) tackles the question of stability. The second contribution is that it finds evidence of an important limitation of the Minimum Distance index when dealing with high dimensional data - the index may become highly concentrated and may remain sub-optimal at all dimensions. Selectively removing noise components by visual inspection can improve the Amari, the Icasso or the MD index values. Some automated tools exist, but in high dimensions, visual inspection of the individual components is still required for effective denoising - data driven methods are not good enough.

Web visualizations have become an integral tool for representing data in an accessible and intelligible format. This paper presents a new workflow, Quick Vis, which enables an analyst to inspect data sets and manipulate properties within any data set. This subsequently engages our analytic engine, saves each property set as a profile (discussed in Section 5), and allows comparisons between profiles for further findings. In addition to describing our proposed techniques, we provide a case study that demonstrates the usefulness of Quick Vis for real life problems.

It has been long thought that the main cause of a heart attack is the narrowing of an artery from the buildup of fatty plaque inside the artery wall. Postmortem autopsy studies on patients who have died after a myocardial infarction (heart attack) found that many of those did not have their arteries severely narrowed by plaque. Instead, thrombosis (blood clots)near ruptured lesions were identified as the actual main cause. The search for the type of culprit lesions has established that the formation of a lipid-rich pool covered by a thin fibrous cap(thickness less than 65 micrometer) was structurally unstable and likely to rupture. Subsequently, the term vulnerable plaque has been used to refer to this type of plaque. Optical Coherence Tomography (OCT) is a new imaging modality capable of providing the required micrometer scale resolution to detect such tissue. With a resolution of approximately15 micrometer, OCT is considered the most promising imaging modality for identifying vulnerable plaque. We present a complete framework for reliable tissue characterization of different plaque types using OCT images. Our framework utilizes texture analysis as delineating features. In addition, a parametric classification technique is presented. The characterization performance of the proposed framework is assessed and evaluated.

The problem of projecting multidimensional data into lower dimensions has been pursued by many researchers due to its potential application to data analysis of various kinds. This paper presents a novel multidimensional projection technique based on least square approximations. The approximations compute the coordinates of a set of projected points based on the coordinates of a reduced number of control points with defined geometry. We name the technique Least Square Projections (LSP). From an initial projection of the control points, LSP defines the positioning of their neighboring points through a numerical solution that aims at preserving a similarity relationship between the points given by a metric in mD. In order to perform the projection, a small number of distance calculations is necessary and no repositioning of the points is required to obtain a final solution with satisfactory precision. The results show the capability of the technique to form groups of points by degree of similarity in 2D. We illustrate that capability through its application to mapping collections of textual documents from varied sources, a strategic yet difficult application. LSP is faster and more accurate than other existing high quality methods, particularly where it was mostly tested, that is, for mapping text sets.

Researchers and users are well aware of the difficulties related to finding an appropriate configuration of the axes mapping attributes in multidimensional visualization techniques, particularly in visualizations that show a large number of attributes simultaneously. We address this problem with a simple strategy that offers both dimension ordering and dimension reduction. Dimension ordering is based on attribute similarity heuristics, and the basic rationale is extended to support dimension reduction. We discuss the performance of our algorithms and present some results of their application to several data sets. The algorithms improve the capability of visualization techniques to segregate clusters present in the data and reduce the visual clutter aggravated by arbitrary distributions of the axes

A set of user evaluation studies were performed to test hypotheses about the efficacy and value of the use of sonification by itself and together with visualization to improve the process of interactive data exploration. Our results show that, in some cases, sonification yielded more accurate analysis results than visualization, and when combined with visualization, yield much better results than either one (visualization or sonification) yielded separately. That supports sonification as a valuable and promising tool that should be further studied as a complement to visualization systems.

We survey work on the different uses of graphical mapping and interaction techniques for visual data mining of large data sets represented as table data. Basic terminology related to data mining, data sets, and visualization is introduced. Previous work on information visualization is reviewed in light of different categorizations of techniques and systems. The role of interaction techniques is discussed, in addition to work addressing the question of selecting and evaluating visualization techniques. We review some representative work on the use of information visualization techniques in the context of mining data. This includes both visual data exploration and visually expressing the outcome of specific mining algorithms. We also review recent innovative approaches that attempt to integrate visualization into the DM/KDD process, using it to enhance user interaction and comprehension.

We present a novel approach to Automatic Language Identification (LID). We propose Perceptually Guided Training (PGT), a novel LID training method, involving identification of utterance parts which are particularly significant perceptually for the language identification process, and exploitation of these Perceptually Significant Regions (PSRs) to guide the LID training process. Our approach involves a Recurrent Neural Network (RNN) as the main mechanism. We propose that, because of the long-range intrautterance acoustical context significance in LID, RNNs are particularly suitable for the LID task. Our approach does not require phonetic labeling or transcription of the training corpus. LIREN/PGT, the LID system we developed, incorporates our approach. Our LID experiments were on English, German, and Mandarin Chinese, using the OGI-TS corpus.

A technique that harnesses color and texture perception to create integrated displays of 2D image-like multiparameter distributions is presented. The power of the technique is demonstrated by an example of a synthesized dataset and compared with several other proposed techniques. The nature of studies that are required to measure objectively and accurately the effectiveness of such displays is discussed.>