Search

Hierarchical clustering allows better performance in grouping heterogeneous and non-spherical datasets than the center-based clustering, at the expense of increased time complexity. Meanwhile, the bottom-up approach of hierarchical clustering methods often tend to be sensitive or vulnerable to datasets containing obscure cluster boundaries. This paper presents an effective method for hierarchical clustering, called HCNN, which utilizes two types of structural similarities in nearest neighbor graph of a dataset to group similar data into clusters. In particular, the first metric is used to identify those pairs of data with maximal similarity in their nearest neighborhoods that can serve as local centers of every initial cluster. This can contribute to observing the boundaries and detecting clusters, hubs and outliers, thereby alleviating remarkably the influence of obscure boundaries between clusters. The initial clusters will be merged recursively in accordance with the second similarity metric measured in terms of the connectivity between clusters in the nearest neighbor graph. In this case, rather than combining two clusters with highest similarity during each iteration, we consider the maximum similarity as a transitive and closure relation between clusters, i.e., an equivalence relation, which enables more effective and efficient merging of clusters via application of advanced data structures. Experiments based on synthetic and real datasets demonstrate that the proposed HCNN algorithms can possibly outperform the state-of-the-art clustering methods evaluated in terms of the clustering accuracy, normalized mutual information and adjust rand index. Moreover, experiments on unlabeled real datasets show that our method enables effective identification of the quantity of clusters based on Davies–Bouldin index and average silhouette coefficient.

A Brownian Cellular Automaton (BCA) is an Asynchronous Cellular Automaton (ACA) in which the timing of the local cell state transitions approaches a Poisson point process, and the global transitions between cell configurations can be formulated as an irreducible and reversible Markov chain. BCAs exploit fluctuations to find solutions of a computation via a stochastic search process. This characteristic has been shown to contribute to a decreased complexity of BCAs (Lee and Peper, 2008), as compared to non-Brownian ACAs with equivalent functionality. This paper proposes a computationally universal BCA that is based on a conventional ACA, rather than the previously employed block cellular automata, which have transition rules that update each cell simultaneously with its neighboring cells. The proposed allows for a more natural update model, in which no need arises to locally synchronize the cells in a neighborhood. The resulting BCA is significantly less complex than equivalent non-block non-Brownian ACAs, which demonstrates that random fluctuations can serve as a powerful resource for efficient computation on cellular automata.

a b s t r a c t We demonstrate optical excitation transfer in a mixture composed of quantum dots of two different sizes (larger and smaller) networked via optical near-field interactions. For the optical near-field interaction network based on a density matrix formalism, we introduce an optimal mixture that agrees with experimental results. Based on these findings, we theoretically examine the topology-dependent efficiency of optical excitation transfer, which clearly exhibits autonomous, energy-efficient networking behavior occurring at the nanometer scale. We discuss what we can learn from this optical excitation transfer and its implications for information and communications applications.

Summary Data applications are expected to play an increasingly important role in the next generation mobile communication services. To plan these networks, detailed models of data users with diverse applications are required. The purpose of this study is to provide practically usable traffic models for data users in wireless networks, e.g. GPRS. We start with a discussion of well-known models in the literature, where the focus is on their applicability in mobile networks. A practical model of mobile HTTP users, based on an extrapolation of WWW users in current wire-line networks, is presented; this model is expected to be accurate enough to be used in simulation studies for network planning purposes. Finally, models for other emerging data applications are also taken into account.