This paper presents an optimal load balancing algorithm based on both of the ANFIS (Adaptive Neuro-Fuzzy Inference System) modeling and the FIS (Fuzzy Inference System) for the local status of real servers. It also shows the substantial benefits such as the removal of load-scheduling overhead, QoS (Quality of Service) provisioning and providing highly available servers, provided by the suggested method.

In this paper, conventional validity indexes are reviewed and the shortcomings of the fuzzy cluster validation index based on inter-cluster proximity are examined. Based on these considerations, a new cluster validity index is proposed for fuzzy partitions obtained from the fuzzy c-means algorithm. The proposed validity index is defined as the average value of the relative intersections of all possible pairs of fuzzy clusters in the system. It computes the overlap between two fuzzy clusters by considering the intersection of each data point in the overlap. The optimal number of clusters is obtained by minimizing the validity index with respect to c. Experiments in which the proposed validity index and several conventional validity indexes were applied to well known data sets highlight the superior qualities of the proposed index.

For future high-speed wireless communications using orthogonal frequency division multiplexing (OFDM), two major system requirements will emerge: throughput improvement and rich interference elimination. Because of its broadband nature and limited frequency allocations worldwide, interference from co-located wireless LAN's operating in the same frequency band will become a serious deployment issue. Adaptive array antenna can enhance the performance by suppressing the co-channel interference even when interference may have a large amount of multipath and also have similar received power to the desired signal. There are typically two types of adaptive array architecture for OFDM systems, whose signal processing is carried out before or after FFT (Fast Fourier Transform). In general, the pre-FFT array processing has low complexity, but in rich multipath and interference environments, the performance will deteriorate drastically. In contrast, the post-FFT array processing can provide the optimum performance even in such severe environments at the cost of complexity. Therefore, complexity-reduction techniques combined with the achievement of high system performance will be a key issue for adaptive array antenna applications. This paper proposes novel adaptive array architecture, which is a complexity-reduction technique using subcarrier clustering for post-FFT adaptive array. In the proposed scheme, plural subcarriers can be clustered into a group with the same spatial weight. Simulation results show that the proposed architecture is a promising candidate for real implementation, since it can achieve high performance with much lower complexity even in a rich multipath environment with low signal to noise plus interference ratio (SNIR).

This paper presents a novel digital image watermarking algorithm based on the labeled bisecting clustering technique. Each cluster is labeled either '0' or '1' based on the labeling key. Each input image block is then assigned to the nearest codeword or cluster centre whose label is equal to the watermark bit. The watermark extraction can be performed blindly. The proposed method is robust to JPEG compression and some spatial-domain processing operations. Simulation results demonstrate the effectiveness of the proposed algorithm.

An approach is proposed for constructing a dependable server cluster composed only of server nodes with all nodes running the same algorithm. The cluster propagates an IP multicast address as the server address, and clients multicast requests to the cluster. A local proxy running on each client machine enables conventional client software designed for unicasting to communicate with the cluster without having to be modified. Evaluation of a prototype system providing domain name service showed that a cluster using this technique has high dependability with acceptable performance degradation.

Automatic text summarization has the goal of reducing the size of a document while preserving its content. Generally, producing a summary as extracts is achieved by including only sentences which are the most topic-related. DOCUSUM is our summarization system based on a new topic keyword identification method. The process of DOCUSUM is as follows. First, DOCUSUM converts the content words of a document into elements of a context vector space. It then constructs lexical clusters from the context vector space and identifies core clusters. Next, it selects topic keywords from the core clusters. Finally, it generates a summary of the document using the topic keywords. In the experiments on various compression ratios (the compression of 30%, the compression of 10%, and the extraction of the fixed number of sentences: 4 or 8 sentences), DOCUSUM showed better performance than other methods.

This paper describes a new training method of average voice model for speech synthesis in which arbitrary speaker's voice is generated based on speaker adaptation. When the amount of training data is limited, the distributions of average voice model often have bias depending on speaker and/or gender and this will degrade the quality of synthetic speech. In the proposed method, to reduce the influence of speaker dependence, we incorporate a context clustering technique called shared decision tree context clustering and speaker adaptive training into the training procedure of average voice model. From the results of subjective tests, we show that the average voice model trained using the proposed method generates more natural sounding speech than the conventional average voice model. Moreover, it is shown that voice characteristics and prosodic features of synthetic speech generated from the adapted model using the proposed method are closer to the target speaker than the conventional method.

The alternative c-means algorithm has recently been presented by Wu and Yang for robust clustering of data. In this letter, we analyze the convergence of this algorithm by transforming it into an equivalent form with the Legendre transform. It is shown that this algorithm converges to a local optimal solution from any starting point.

This paper evaluates an on-line incremental speaker adaptation method for co-channel conversation including multiple speakers with the assumption that the speaker is unknown and changes frequently. After performing the speaker clustering treatment based on the Vector Quantization (VQ) distortion for every utterance, acoustic models for each cluster are adapted by Maximum Likelihood Linear Regression (MLLR) or Maximum A Posteriori probability (MAP). The performance of continuous speech recognition could be improved. In this paper, to prove the efficiency of the speaker clustering method for improving the performance of continuous speech recognition, the continuous speech recognition experiments with supervised and unsupervised cluster adaptation were conducted, respectively. Finally, evaluation experiments based on other prepared test data were performed on continuous syllable recognition and large vocabulary continuous speech recognition (LVCSR). The efficiency of the speaker adaptation and clustering methods presented in this paper was supported strongly by the experimental results.

This paper describes a new context clustering technique for average voice model, which is a set of speaker independent speech synthesis units. In the technique, we first train speaker dependent models using multi-speaker speech database, and then construct a decision tree common to these speaker dependent models for context clustering. When a node of the decision tree is split, only the context related questions which are applicable to all speaker dependent models are adopted. As a result, every node of the decision tree always has training data of all speakers. After construction of the decision tree, all speaker dependent models are clustered using the common decision tree and a speaker independent model, i.e., an average voice model is obtained by combining speaker dependent models. From the results of subjective tests, we show that the average voice models trained using the proposed technique can generate more natural sounding speech than the conventional average voice models.

In this paper, a new voice personality transformation algorithm which uses the vocal tract characteristics and pitch period as feature parameters is proposed. The vocal tract transfer function is divided into time-invariant and time-varying parts. Conversion rules for the time-varying part are constructed by the classified-linear transformation matrix based on soft-clustering techniques for LPC cepstrum expressed in KL (Karhunen-Loève) coefficients. An excitation signal containing prosodic information is transformed by average pitch ratio. In order to improve the naturalness, transformation on the excitation signal is separately applied to voiced and unvoiced bands to preserve the overall spectral structure. Objective tests show that the distance between the LPC cepstrum of a target speaker and that of the speech synthesized using the proposed method is reduced by about 70% compared with the distance between the target speaker's LPC cepstrum and the source speaker's. Also, subjective listening tests show that 60-70% of listeners identify the transformed speech as the target speaker's.

This paper discusses a new type of semi-supervised document clustering that uses partial supervision to partition a large set of documents. Most clustering methods organizes documents into groups based only on similarity measures. In this paper, we attempt to isolate more semantically coherent clusters by employing the domain-specific knowledge provided by a document analyst. By using external human knowledge to guide the clustering mechanism with some flexibility when creating the clusters, clustering efficiency can be considerably enhanced. Experimental results show that the use of only a little external knowledge can considerably enhance the quality of clustering results that satisfy users' constraint.

This paper proposed an automated segmentation algorithm for MR brain images through the complementary use of T1-weighted, T2-weighted, and PD images. The proposed segmentation algorithm is composed of 3 steps. The first step involves the extraction of cerebrum images by placing a cerebrum mask over the three input images. In the second step, outstanding clusters that represent the inner tissues of the cerebrum are chosen from among the 3-dimensional (3D) clusters. The 3D clusters are determined by intersecting densely distributed parts of a 2D histogram in 3D space formed using three optimal scale images. The optimal scale image results from applying scale-space filtering to each 2D histogram and a searching graph structure. As a result, the optimal scale image can accurately describe the shape of the densely distributed pixel parts in the 2D histogram. In the final step, the cerebrum images are segmented by the FCM (Fuzzy c-means) algorithm using the outstanding cluster center value as the initial center value. The ability of the proposed segmentation algorithm to calculate the cluster center value accurately then compensates for the current limitation of the FCM algorithm, which is unduly restricted by the initial center value used. In addition, the proposed algorithm, which includes a multi spectral analysis, can achieve better segmentation results than a single spectral analysis.

Clustering is one of the most important topics in the field of knowledge discovery from databases. Especially, hierarchical clustering is useful since it gives a hierarchical view of a whole database and can be used to guide users in browsing a huge database. In many cases, clustering can be modeled as a graph partitioning problem. When an appropriate distance function between database objects is given, a database can be viewed as an edge-weighted complete graph, where vertices are database objects and weights of edges are distances between them. Then a process of MST (Minimal Spanning Tree) construction can be viewed as a process of a single-linkage agglomerative clustering process for database objects. In this paper, we propose an efficient MST construction method for a large complete metric graph, which is derived from a database with a metric distance function defined on it. Our method utilizes a metric index to reduce the number of distance calculations. The basic idea is to exclude those edges less probable to be a part of an MST by using the metric postulate. For this purpose, we introduce a new metric index named MetricMatrix. Experimental results show that our method can drastically reduce the number of distance calculations needed for MST construction in comparison with the classical method.

Most scheduling applications have been classified into NP-complete problems. This fact implies that an optimal solution for a large scheduling problem is extremely time-consuming. A number of schemes are introduced to solve NP-complete scheduling applications, such as linear programming, neural network, and fuzzy logic. In this paper, we demonstrate a new approach, fuzzy Hopfield neural network, to solve the scheduling problems. This fuzzy Hopfield neural network approach integrates fuzzy c-means clustering strategies into a Hopfield neural network. In this investigation, we utilizes this new approach to demonstrate the feasibility of resolving a multiprocessor scheduling problem with no process migration, limited resources and constrained times (execution time and deadline). In the approach, the process and processor of the scheduling problem can be regarded as a data sample and a cluster, respectively. Then, an appropriate Lyapunov energy function is derived correspondingly. The scheduling results can be obtained using a fuzzy Hopfield neural network clustering technique by iteratively updating fuzzy state until the energy function gets minimized. To validate our approach, three scheduling cases for different initial neuron states as well as fuzzification parameters are taken as testbed. Simulation results reveal that imposing the fuzzy Hopfield neural network on the proposed energy function provides a sound approach in solving this class of scheduling problems.

In this paper we consider the k-clustering problem for a set S of n points pi=(xi) in the d-dimensional space with variance-based errors as clustering criteria, motivated from the color quantization problem of computing a color lookup table for frame buffer display. As the inter-cluster criterion to minimize, the sum of intra-cluster errors over every cluster is used, and as the intra-cluster criterion of a cluster Sj, |Sj|α-1 Σpi Sj ||xi - - x (Sj)||2 is considered, where |||| is the L2 norm and - x (Sj) is the centroid of points in Sj, i. e. , (1/|Sj|)Σpi Sj xi. The cases of α=1,2 correspond to the sum of squared errors and the all-pairs sum of squared errors, respectively. The k-clustering problem under the criterion with α = 1,2 are treated in a unified manner by characterizing the optimum solution to the k-clustering problem by the ordinary Euclidean Voronoi diagram and the weighted Voronoi diagram with both multiplicative and additive weights. With this framework, the problem is related to the generalized primary shatter function for the Voronoi diagrams. The primary shatter function is shown to be O(nO(kd)), which implies that, for fixed k, this clustering problem can be solved in a polynomial time. For the problem with the most typical intra-cluster criterion of the sum of squared errors, we also present an efficient randomized algorithm which, roughly speaking, finds an ε-approximate 2-clustering in O(n(1/ε)d) time, which is quite practical and may be used to real large-scale problems such as the color quantization problem.

This paper surveys how geometric information can be effectively used for efficient algorithms with focus on clustering problems. Given a complete weighted graph G of n vertices, is there a partition of the vertex set into k disjoint subsets so that the maximum weight of an innercluster edge (whose two endpoints both belong to the same subset) is minimized? This problem is known to be NP-complete even for k = 3. The case of k=2, that is, bipartition problem is solvable in polynomial time. On the other hand, in geometric setting where vertices are points in the plane and weights of edges equal the distances between corresponding points, the same problem is solvable in polynomial time even for k 3 as far as k is a fixed constant. For the case k=2, effective use of geometric property of an optimal solution leads to considerable improvement on the computational complexity. Other related topics are also discussed.

This paper surveys recent progress in the investigation of the underlying discrete proximity structures of geometric clustering with respect to the divergence in information geometry. Geometric clustering with respect to the divergence provides powerful unsupervised learning algorithms, and can be applied to classifying and obtaining generalizations of complex objects represented in the feature space. The proximity relation, defined by the Voronoi diagram by the divergence, plays an important role in the design and analysis of such algorithms.

In this paper, a clustering-based method is proposed for automatically constructing a multi-input Takagi-Sugeno (TS) fuzzy model where only the input-output data of the identified system are available. The TS fuzzy model is automatically generated by the process of structure identification and parameter identification. In the structure identification step, a clustering method is proposed to provide a systematic procedure to partition the input space so that the number of fuzzy rules and the shapes of fuzzy sets in the premise part are determined from the given input-output data. In the parameter identification step, the recursive least-squares algorithm is applied to choose the parameter values in the consequent part from the given input-output data. Finally, two examples are used to illustrate the effectiveness of the proposed method.

In this paper, a novel variable-rate vector quantizer (VQ) design algorithm using fuzzy clustering technique is presented. The algorithm, termed fuzzy entropy-constrained VQ (FECVQ) design algorithm, has a better rate-distortion performance than that of the usual entropy-constrained VQ (ECVQ) algorithm for variable-rate VQ design. When performing the fuzzy clustering, the FECVQ algorithm considers both the usual squared-distance measure, and the length of channel index associated with each codeword so that the average rate of the VQ can be controlled. In addition, the membership function for achieving the optimal clustering for the design of FECVQ are derived. Simulation results demonstrate that the FECVQ can be an effective alternative for the design of variable-rate VQs.