In the last years, well-ordered nanoclusters attract a lot of attention, because they are effectively used to enhance the catalytic capability and Curie temperature for ferromagnetism. However, in the conventional case, the nucleation sites will form 2-dimensional islands and then grow into films with increasing coverage. This kind of 2-D islands restricts the function for the further applications. Thus, in this report, we controlled the growth of 3-D Fe nanoclusters by the method of buffer layer assisted growth (BLAG) on Au(111) herringbone reconstruction structure. The method of BLAG was used to control the Fe nanoclusters size from 2 to 6 nm by the amount of Xe buffer layer and Fe deposition. In addition, the Fe nanodots are confined and arranged at the turning points of the Au(111) herringbone by Fe seeds before the BLAG method. They can be used as the nucleation sites to assemble the well-order nanoclusters by BLAG. From the result, the size controlled 3-D Fe nanoclusters self-assemble at special point. The method of ¡§BLAG on nano-patterned template¡¨ will be very helpful to prepare various 3-D nanoclusters with regular spatial arrangement and to control size of them.

Thesis (M. Phil.)--University of Hong Kong, 2010. / Includes bibliographical references (leaves 61-66). Also available in print.

Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (p. 132-140).

Zugl.: Berlin, Freie Universiẗat, Diss., 2009.

Zugl.: Karlsruhe, Univ., Diss., 2009 / Hergestellt on demand. - Zusätzliches Online-Angebot unter http://uvka.ubka.uni-karlsruhe.de/shop/isbn/978-3-86644-392-1

No description available.

Co-clustering is rather a recent paradigm for unsupervised data analysis, but it has become increasingly popular because of its potential to discover latent local patterns, otherwise unapparent by usual unsupervised algorithms such as k-means. Wide deployment of co-clustering, however, requires addressing a number of practical challenges such as data transformation, cluster initialization, scalability, and so on. Therefore, this thesis focuses on developing sophisticated co-clustering methodologies to maturity and its ultimate goal is to promote co-clustering as an invaluable and indispensable unsupervised analysis tool for varied practical applications. To achieve this goal, we explore the three specific tasks: (1) development of co-clustering algorithms to be functional, adaptable, and scalable (co-clustering algorithms); (2) extension of co-clustering algorithms to incorporate application-specific requirements (extensions); and (3) application of co-clustering algorithms broadly to existing and emerging problems in practical application domains (applications). As for co-clustering algorithms, we develop two fast Minimum Sum-Squared Residue Co-clustering (MSSRCC) algorithms [CDGS04], which simultaneously cluster data points and features via an alternating minimization scheme and generate co-clusters in a “checkerboard” structure. The first captures co-clusters with constant values, while the other discovers co-clusters with coherent “trends” as well as constant values. We note that the proposed algorithms are two special cases (bases 2 and 6 with Euclidean distance, respectively) of the general co-clustering framework, Bregman Co-clustering (BCC) [BDG+07], which contains six Euclidean BCC and six I-divergence BCC algorithms. Then, we substantially enhance the performance of the two MSSRCC algorithms by escaping from poor local minima and resolving the degeneracy problem of generating empty clusters in partitional clustering algorithms through the three specific strategies: (1) data transformation; (2) deterministic spectral initialization; and (3) local search strategy. Concerning co-clustering extensions, we investigate general algorithmic strategies for the general BCC framework, since it is applicable to a large class of distance measures and data types. We first formalize various data transformations for datasets with varied scaling and shifting factors, mathematically justify their effects on the six Euclidean BCC algorithms, and empirically validate the analysis results. We also adapt the local search strategy, initially developed for the two MSSRCC algorithms, to all the twelve BCC algorithms. Moreover, we consider variations of cluster assignments and cluster updates, including greedy vs. non-greedy cluster assignment, online vs. batch cluster update, and so on. Furthermore, in order to provide better scalability and usability, we parallelize all the twelve BCC algorithms, which are capable of co-clustering large-scaled datasets over multiple processors. Regarding co-clustering applications, we extend the functionality of BCC to incorporate application-specific requirements: (1) discovery of inverted patterns, whose goal is to find anti-correlation; (2) discovery of coherent co-clusters from noisy data, whose purpose is to do dimensional reduction and feature selection; and (3) discovery of patterns from time-series data, whose motive is to guarantee critical time-locality. Furthermore, we employ co-clustering to pervasive computing for mobile devices, where the task is to extract latent patterns from usage logs as well as to recognize specific situations of mobile-device users. Finally, we demonstrate the applicability of our proposed algorithms for aforementioned applications through empirical results on various synthetic and real-world datasets. In summary, we present co-clustering algorithms to discover latent local patterns, propose their algorithmic extensions to incorporate specific requirements, and provide their applications to a wide range of practical domains. / text

published_or_final_version / Physics / Master / Master of Philosophy

The renormalization group (RG) approach is largely responsible for the considerable success which has been achieved in developing a quantitative theory of phase transitions. This work investigates various spectral properties of the RG map for Ising-type classical lattice systems. It consists of four parts. The first part carries out some explicit calculations of the spectrum of the linearization of the RG at infinite temperature, and discovers that it is of an unusual kind: dense point spectrum for which the adjoint operators have no point spectrum at all, but only residual spectrum. The second part presents a rigorous justification of the existence and differentiability of the RG map in the infinite volume limit at high temperature by a cluster expansion approach. The third part continues the theme of the third part, and shows that the matrix of partial derivatives of the RG map displays an approximate band property for finite-range and translation-invariant Hamiltonians at high temperature. The last part justifies the differentiability of the RG map in the infinite volume limit at the critical temperature under a certain condition. In summary, the first part deals with special cases where exact computations can be done, whereas the remaining parts are concerned with a general theory and provide a mathematically sound base.

We propose a novel framework based on Bayesian principles for validating clusterings and present efficient algorithms for use with centroid or exemplar based clustering solutions. Our framework treats the data as fixed and introduces perturbations into the clustering procedure. In our algorithms, we scale the distances between points by a random variable whose distribution is tuned against a baseline null dataset. The random variable is integrated out, yielding a soft assignment matrix that gives the behavior under perturbation of the points relative to each of the clusters. From this soft assignment matrix, we are able to visualize inter-cluster behavior, rank clusters, and give a scalar index of the the clustering stability. In a large test on synthetic data, our method matches or outperforms other leading methods at predicting the correct number of clusters. We also present a theoretical analysis of our approach, which suggests that it is useful for high dimensional data.