Infinite factorization of multiple non-parametric views

Combined analysis of multiple data sources has increasing application interest, in particular for distinguishing shared and source-specific aspects. We extend this rationale to the generative and non-parametric clustering setting by introducing a novel non-parametric hierarchical mixture model. The lower level of the model describes each source with a flexible non-parametric mixture, and the top level combines these to describe commonalities of the sources. The lower-level clusters arise from hierarchical Dirichlet Processes, inducing an infinite-dimensional contingency table between the sources. The commonalities between the sources are modeled by an infinite component model of the contingency table, interpretable as non-negative factorization of infinite matrices, or as a prior for infinite contingency tables. With Gaussian mixture components plugged in for continuous measurements, the model is applied to two views of genes, mRNA expression and abundance of the produced proteins, to expose groups of genes that are co-regulated in either or both of the views. We discover complex relationships between the marginals (that are multimodal in both marginals) that would remain undetected by simpler models. Cluster analysis of co-expression is a standard method of screening for co-regulation, and the two-view analysis extends the approach to distinguishing between pre- and post-translational regulation.     

Three-dimensional interactive Molecular Dynamics program for the study of defect dynamics in crystals

The study of crystal defects and the complex processes underlying their formation and time evolution has motivated the development of the program ALINE for interactive molecular dynamics experiments. This program couples a molecular dynamics code to a Graphical User Interface and runs on a UNIX-X11 Window System platform with the MOTIF library, which is contained in many standard Linux releases. ALINE is written in C, thus giving the user the possibility to modify the source code, and, at the same time, provides an effective and user-friendly framework for numerical experiments, in which the main parameters can be interactively varied and the system visualized in various ways. We illustrate the main features of the program through some examples of detection and dynamical tracking of point-defects, linear defects, and planar defects, such as stacking faults in lattice-mismatched heterostructures.

Program summary

Title of program:ALINECatalogue identifier:ADYJ_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/ADYJ_v1_0Program obtainable from: CPC Program Library, Queen University of Belfast, N. IrelandComputer for which the program is designed and others on which it has been tested: Computers:DEC ALPHA 300, Intel i386 compatible computers, G4 Apple ComputersInstallations:Laboratory of Computational Engineering, Helsinki University of Technology, Helsinki, FinlandOperating systems under which the program has been tested:True64 UNIX, Linux-i386, Mac OS X 10.3 and 10.4Programming language used:Standard C and MOTIF librariesMemory required to execute with typical data:6 Mbytes but may be larger depending on the system sizeNo. of lines in distributed program, including test data, etc.:16 901No. of bytes in distributed program, including test data, etc.:449 559Distribution format:tar.gzNature of physical problem:Some phenomena involving defects take place inside three-dimensional crystals at times which can be hardly predicted. For this reason they are difficult to detect and track even within numerical experiments, especially when one is interested in studying their dynamical properties and time evolution. Furthermore, traditional simulation methods require the storage of a huge amount of data which in turn may imply a long work for their analysis.Method of solution:Simplifications of the simulation work described above strongly depend also on the computer performance. It has now become possible to realize some of such simplifications thanks to the real possibility of using interactive programs. The solution proposed here is based on the development of an interactive graphical simulation program both for avoiding large storage of data and the subsequent elaboration and analysis as well as for visualizing and tracking many phenomena inside three-dimensional samples. However, the full computational power of traditional simulation programs may not be available in general in programs with graphical user interfaces, due to their interactive nature. Nevertheless interactive programs can still be very useful for detecting processes difficult to visualize, restricting the range or making a fine tuning of the parameters, and tailoring the faster programs toward precise targets.Restrictions on the complexity of the problem:The restrictions on the applicability of the program are related to the computer resources available. The graphical interface and interactivity demand computational resources that depend on the particular numerical simulation to be performed. To preserve a balance between speed and resources, the choice of the number of atoms to be simulated is critical. With an average current computer, simulations of systems with more than 10⁵ atoms may not be easily feasible on an interactive scheme. Another restriction is related to the fact that the program was originally designed to simulate systems in the solid phase, so that problems in the simulation may occur if some particular physical quantities are computed beyond the melting point.Typical running time:It depends on the machine architecture, system size, and user needs.Unusual features of the program:In the program, besides the window in which the system is represented in real space, an additional graphical window presenting the real time distribution histogram for different physical variables (such as kinetic or potential energy) is included. Such tool is very interesting for making demonstrative numerical experiments for teaching purposes as well as for research, e.g., for detecting and tracking crystal defects. The program includes: an initial condition builder, an interactive display of the simulation, a set of tools which allow the user to filter through different physical quantities the information—either displayed in real time or printed in the output files—and to perform an efficient search of the interesting regions of parameter space.     

Self-organizing map in recognition of topographic patterns of EEGspectra

The self-organizing map, a neural network algorithm, was applied to the recognition of topographic patterns in clinical 22-channel EEG. Inputs to the map were extracted from short-time power spectra of all channels. Each location on a self-organized map entails a model for a cluster of similar input patterns; the best-matching model determines the location of a sample on the map. Thus, an instantaneous topographic EEG pattern corresponds to the location of the sample, and changes with time correspond to the trajectories of consecutive samples. EEG segments of “alpha”, “alpha attenuation”, “theta of drowsiness”, “eye movements”, “EMG artifact”, and “electrode artifacts” were selected and labeled by visual inspection of the original records. The map locations of the labeled segments were different; the map thus distinguished between them. The locations of individual EEG's on the “alpha-area” of the map were also different. The clustering and easily understandable visualization of topographic EEG patterns are obtainable on a self-organized map in real time     

Simulations of crack initiation in silicon

The applicability of semiempirical potential energy models for describing crack initiation in covalently bonded silicon has been studied using classical molecular dynamics (MD) approach. For describing interatomic interactions, in this approach we use the recently developed environment dependent interatomic potential (EDIP) with two- and three-body terms. Since the original form of this potential was found problematic in describing bond-breaking properties we tested three different modifications of it. An additional point of interest in this study were crack tip structures observed preceding the actual fracture. Our results, with an idealized simulation setup, indicated formation of stable ring-like structures. Unless angular forces were made relatively strong, these ring-like structures were formed near the crack tip before and even during the crack initiation. These relatively stable structures could cause crack initiation to stop temporarily, especially at early stages of fracture.     

Modelling opinion formation driven communities in social networks

In a previous paper we proposed a model to study the dynamics of opinion formation in human societies by a co-evolution process involving two distinct time scales of fast transaction and slower network evolution dynamics. In the transaction dynamics we take into account short range interactions as discussions between individuals and long range interactions to describe the attitude to the overall mood of society. The latter is handled by a uniformly distributed parameter α, assigned randomly to each individual, as quenched personal bias. The network evolution dynamics is realised by rewiring the societal network due to state variable changes as a result of transaction dynamics. The main consequence of this complex dynamics is that communities emerge in the social network for a range of values in the ratio between time scales. In this paper we focus our attention on the attitude parameter α and its influence on the conformation of opinion and the size of the resulting communities. We present numerical studies and extract interesting features of the model that can be interpreted in terms of social behaviour.     

Principle of Learning Metrics for Exploratory Data Analysis

Visualization and clustering of multivariate data are usually based on mutual distances of samples, measured by heuristic means such as the Euclidean distance of vectors of extracted features. Our recently developed methods remove this arbitrariness by learning to measure important differences. The effect is equivalent to changing the metric of the data space. It is assumed that variation of the data is important only to the extent it causes variation in auxiliary data which is available paired to the primary data. The learning of the metric is supervised by the auxiliary data, whereas the data analysis in the new metric is unsupervised. We review two approaches: a clustering algorithm and another that is based on an explicitly generated metric. Applications have so far been in exploratory analysis of texts, gene function, and bankruptcy. Relationships of the two approaches are derived, which leads to new promising approaches to the clustering problem.