Chi-Kwong Li

Dissimilarity learning for nominal data

Defining a good distance (dissimilarity) measure between patterns is of crucial importance in many classification and clustering algorithms. While a lot of work has been performed on continuous attributes, nominal attributes are more difficult to handle. A popular approach is to use the value difference metric (VDM) to define a real-valued distance measure on nominal values. However, VDM treats the attributes separately and ignores any possible interactions among attributes. In this paper, we propose the use of adaptive dissimilarity matrices for measuring the dissimilarities between nominal values. These matrices are learned via optimizing an error function on the training samples. Experimental results show that this approach leads to better classification performance. Moreover, it also allows easier interpretation of (dis)similarity between different nominal values.

Gaussian Mixture Models and Probabilistic Decision-Based Neural Networks for Pattern Classification: A Comparative Study

Probabilistic Decision-Based Neural Networks (PDBNNs) can be considered as a special form of Gaussian Mixture Models (GMMs) with trainable decision thresholds. This paper provides detailed illustrations to compare the recognition accuracy and decision boundaries of PDBNNs with that of GMMs through two pattern recognition tasks, namely the noisy XOR problem and the classification of two-dimensional vowel data. The paper highlights the strengths of PDBNNs by demonstrating that their thresholding mechanism is very effective in detecting data not belonging to any known classes. The original PDBNNs use elliptical basis functions with diagonal covariance matrices, which may be inappropriate for modelling feature vectors with correlated components. This paper overcomes this limitation by using full covariance matrices, and showing that the matrices are effective in characterising non-spherical clusters.

Strategic selection and replication of movies by trend-calibrated movie-demand model

To cope with the fluctuation in movie popularity, video-on-demand systems need to maintain an optimal set of movies. Replication of the most popular movies across multiple storage devices is also needed to increase the bandwidth of these movies in meeting their high demand. However, selection and replication of movies can only be varied within the constraints imposed by the online storage space and organization. We evaluate three movie-demand models based on which strategic decisions on selection and replication can be exercised. The long-term static movie-demand (SMD) model is a quantized version of a trivial approach. The bounded-SMD (B-SMD) model is an improved version of SMD with bounded movie lifespan. We propose a new model called the trend-calibrated movie-demand (TCMD) model. Through simulation studies we have found that TCMD can significantly improve the percentage of customer requests satisfied.

Scheduling tasks in DAG to heterogeneous processor system

Heterogeneous processors configuration in parallel and distributed becomes a practical solution in modern parallel and distributed system. In order to execute tasks in such system with better performance, scheduling algorithms which support the configuration are needed. This paper starts with studying a technique called Heterogeneous List Scheduling Heuristic (HLS) for designing scheduling algorithm to schedule tasks into heterogeneous systems. With this, an experience of designing scheduling algorithm for scheduling task into heterogeneous system is described. This is done by modifying an algorithm called Relative Mobility, which is proposed by Chan and Li [2][3] for scheduling task into homogeneous system, to propose an algorithm called Heterogeneous Relative Mobility Scheduling algorithm (HRMS). Finally, an experiment is conducted to show some important properties as scheduling tasks into different configurations of processors.

A fast algorithm for morphological operations with flat structuring element

Flat structuring elements are commonly used in morphological operations. In this paper, a fast algorithm employing the result of previous searching area, which is determined by a domain-selection method, is proposed. It is applicable to structuring elements conforming to a constraint that its one-dimensional (1-D) Euler-Poincare constants, N/sup (1/)(x) and N/sup (1/)(y), at any x- or y-coordinate must be equal to 1. The proposed algorithm is compared with three other methods, namely threshold linear convolution of Kisacanin and Schonfeld (KS), structuring element decomposition of Shih and Mitchell (SM), and fast implementation of Wang and He (WH), in terms of the theoretical expected number of comparisons and experimental computation time. It is found that the proposed algorithm requires less computation time than KS and SM methods for nearly all sizes of square, octagon, and rhombus structuring elements, except for the size of 3/spl times/3. In addition, it is also more time efficient than the WH method, except for the square structuring element.

Heterogeneous Dominant Sequence Cluster (HDSC): a low complexity heterogeneous scheduling algorithm

A novel list scheduling technique, Heterogeneous List Scheduling technique (HLS), is presented. The technique is a procedure to design a scheduling algorithm for scheduling tasks into a heterogeneous environment. Using the technique, a Heterogeneous Dominant Sequence Cluster algorithm (HDSC) is proposed. This algorithm uses the critical paths and dominant sequence as scheduling heuristics with the goal of minimizing the parallel time. It is found that the proposed algorithm is of lower complexity when compared with ordinary designs and it is also suitable for use in a heterogeneous environment. Details of operations and characteristics of the scheduler being tested in different processor configurations have been conducted and satisfactory results are observed.

Elliptical basis function networks and radial basis function networks for speaker verification: a comparative study

It is well known that radial basis function (RBF) networks require a large number of function centers if the data to be modeled contain clusters with complicated shape. This paper proposes to overcome this problem by incorporating full covariance matrices into the RBF structure and to use the expectation-maximization (EM) algorithm to estimate the network parameters. The resulting networks, referred to as the elliptical basis function (EBF) networks, are applied to text-independent speaker verification. Experimental evaluations based on 258 speakers of the TIMIT corpus show that smaller size EBF networks with basis function parameters determined by the EM algorithm outperform the large RBF networks trained by the conventional approach.

Determining the Optimal Number of Clusters by an Extended RPCL Algorithm

Determining an appropriate number of clusters is a diicult yet important problem. The rival penalized competitive learning (RPCL) algorithm was designed to solve this problem. But its performance is not satisfactory when there are overlapped clusters or in the cases where the input vectors contain dependent components. This paper addresses this problem by incorporating full covariance matrices into the original RPCL algorithm. The resulting algorithm, referred to as the extended RPCL algorithm, progressively eliminates the units whose clusters contain only a small portion of the training data. The algorithm is applied to determine the number of clusters of a Gaussian mixture distribution. It is also applied to optimize the architecture of elliptical basis function networks for speaker veriication and for vowel classiication. It is found that the covariance matrices obtained by the extended RPCL algorithm have a better representation of the clusters than that obtained by the original RPCL algorithm, resulting in a lower veriication error rate in the speaker veriication problem and a higher recognition accuracy in the vowel classiication problem.

Maximum likelihood estimation of elliptical basis function parameters with application to speaker verification

The use of the K-means algorithm and the K-nearest neighbor heuristic in estimating the radial basis function (RBF) parameters may produce sub-optimal performance when the input vectors contain correlated components. This paper proposes to overcome this problem by incorporating full covariance matrices into the RBF structure and to use the expectation-maximization (EM) algorithm to estimate the network parameters. The resulting networks, referred to as elliptical basis function (EBF) networks, are applied to text-independent speaker verification. To examine the robustness of the networks in a noisy environment, both clean speech and telephone speech have been used. Experimental results show that smaller size EBF networks with basis function parameters determined by the EM algorithm outperform the large RBF networks trained in the conventional approach. The best error rates achieved by the EBF networks is 3.70%, while that achieved by the RBF networks is 10.37%.

Probabilistic decision-based neural networks for speech pattern classification

Probabilistic decision-based neural networks (PDBNNs) were originally proposed by Lin, Kung and Lin (1997) for human face recognition. Although high recognition accuracy has been achieved, not many illustrations were given to highlight the characteristics of the decision boundaries. This paper aims at providing detailed illustrations to compare the decision boundaries of PDBNNs with that of Gaussian mixture models through a pattern recognition task, namely the classification of two-dimensional vowel data. The original PDBNNs use elliptical basis functions with diagonal covariance matrices, which may be inefficient for modeling feature vectors with correlated components. This paper attempts to tackle this problem by using full covariance matrices. The paper also highlights the strengths of PDBNNs by demonstrating that the PDBNNs thresholding mechanism is very effective in rejecting data not belonging to any known classes.