Feature subset selection is an important preprocessing task for pattern recognition, machine learning or data mining applications. A Genetic Algorithm (GA) with a fuzzy fitness function has been proposed here for finding out the optimal subset of features from a large set of features. Genetic algorithms are robust but time consuming, specially GA with neural classifiers takes a long time for reasonable solution. To reduce the time, a fuzzy measure for evaluation of the quality of a feature subset is used here as the fitness function instead of classifier error rate. The computationally light fuzzy fitness function lowers the computation time of the traditional GA based algorithm with classifier accuracy as the fitness function. Simulation over two data sets shows that the proposed algorithm is efficient for selection of near optimal solution in practical problems specially in case of large feature set problems.

In this work we study the performance of a distributed GA that incorporates in its core parallel cooperative-competitive genetic operators. A series of controlled experiments are conducted using various large and difficult 0/1 multiple knapsack problems to test the robustness of the distributed GA. Simulation results verify that the proposed distributed GA compared with a canonical distributed GA significantly gains in search speed and convergence reliability with less communication cost for migration.

A new blind restoration method applying Real-coded genetic algorithm (RcGA) will be proposed, and this method will be proven valid for the blurred image restoration with unidentified degradation in the experiments. In this restoration method, the degraded and blurred image is going to get restricted to the images possible to be expressed in the point spread function (PSF), then the restoration filter for this degraded image, which is also the 2-dimentional inverse filter, will be searched among several points applying RcGA. The method will enable to seek efficiently among vast solution space consists of numeral coefficient filters. And perceiving the essential features of the spectrum in the frequency space, an evaluation function will be proposed. Also, it will be proposed to apply the Rolling-ball transform succeeding an appropriate Gaussian degrade function against the dual degraded image with blur convoluting impulse noise. By above stated features of this restoration method, it will enable to restore the degraded image closer to the original within a practical processing time. Computer simulations verify this method for image restoration problem when the factors causing image distortions are not identified.

This paper first presents robust algorithms to extract invariants of the linear Lie algebra model from 3D objects. In particular, an extended 3D Hough transform is presented to extract accurate estimates of the normal vectors. The Least square fitting is used to find normal vectors and representation matrices. Then an algorithm of segmentation for 3D objects is shown using the invariants of the linear Lie algebra. Distributions of invariants, both in the invariant space and on the object surface, are used for clustering and edge detection.

Adaptive optimization of the notch bandwidth of a lattice-based adaptive infinite impulse response (IIR) notch filter is presented in this paper. The filter is used to improve the performance of a direct sequence spread spectrum (DSSS) binary phase shift keying (BPSK) communication system by suppressing a narrow-band interference at the receiver. A least mean square (LMS) algorithm used to adapt the notch bandwidth coefficient to its optimum value which corresponds to the maximum signal to noise ratio (SNR) improvement factor is derived. Bit error rate (BER) improvement gained by the DSSS communication system using the filter with the optimized notch bandwidth is also shown. Computer simulation results are compared with those obtained analytically to demonstrate the validity of theoretical predictions for various received signal parameters.

An image halftoning technique that uses a simple GA has proven to be effective generating bi-level halftone images with quality higher than conventional techniques. Many devices are designed to handle more than two halftone levels and a GA based multi-level halftoning technique is desirable. In this paper we extend the bi-level halftoning technique to generate multi-level halftone images. Also we introduce an improved GA (GA-SRM) into the proposed multi-level halftoning technique. Experimental results show that the proposed technique can effectively generate high quality multi-level halftone images and that the inclusion of GA-SRM substantially contributes reducing memory usage and accelerating image generation.

In this paper, we investigate the optimal assignment problem of cells in PCS (Personal Communication Service) to switches on a ATM (Asynchronous Transfer Mode) network. Given cells and switches on an ATM network (whose locations are fixed and known), the problem is to group cells into clusters and assign these clusters to switches in an optimum manner. This problem is modeled as a complex integer programming problem. Since finding an optimal solution of this problem is NP-hard, a heuristic solution model consists of three phases (Cell Pre-Partitioning Phase, Cell Exchanging Phase, and Cell Migrating Phase) is proposed. Experimental results show that Cell Exchanging and Cell Migrating Phases can really reduce total cost near 44% on average.

A plane drawing of a graph is called a floorplan if every face (including the outer face) is a rectangle. A based floorplan is a floorplan with a designated base line segment on the outer face. In this paper we give a simple algorithm to generate all based floorplans with at most n faces. The algorithm uses O(n) space and generates such floorplans in O(1) time per floorplan without duplications. The algorithm does not output entire floorplans but the difference from the previous floorplan. By modifying the algorithm we can generate without duplications all based floorplans having exactly n faces in O(1) time per floorplan. Also we can generate without duplications all (non-based) floorplans having exactly n faces in O(n) time per floorplan.

Software with design flaws increases maintenance costs, decreases component reuse, and reduces software life. Even well-designed software tends to deteriorate with time as it undergoes maintenance. Work on restructuring object-oriented designs involves estimating the quality of the designs using metrics, and automating transformations that preserve the behavior of the designs. However, these factors have been treated almost independently of each other. A long-term goal is to define transformations preserving the behavior of object-oriented designs, and automate the transformations using metrics. In this paper, we describe a genetic algorithm based restructuring approach using metrics to automatically modify object-oriented designs. Cohesion and coupling metrics based on abstract models are defined to quantify designs and provide criteria for comparing alternative designs. The abstract models include a call-use graph and a class-association graph that represent methods, attributes, classes, and their relationships. The metrics include cohesion, inheritance coupling, and interaction coupling based on the behavioral similarity between methods extracted from the models. We define restructuring operations, and show that the operations preserve the behavior of object-oriented designs. We also devise a fitness function using cohesion and coupling metrics, and automatically restructure object-oriented designs by applying a genetic algorithm using the fitness function.

Studies on artificial neural network have been conducted for a long time, and its contribution has been shown in many fields. However, the application of neural networks in the real world domain is still a challenge, since nature does not always provide the required satisfactory conditions. One example is the class size imbalanced condition in which one class is heavily under-represented compared to another class. This condition is often found in the real world domain and presents several difficulties for algorithms that assume the balanced condition of the classes. In this paper, we propose a method for solving problems posed by imbalanced training sets by applying the modified large-scale neural network "CombNET-II. " CombNET-II consists of two types of neural networks. The first type is a one-layer vector quantization neural network to turn the problem into a more balanced condition. The second type consists of several modules of three-layered multilayer perceptron trained by backpropagation for finer classification. CombNET-II combines the two types of neural networks to solve the problem effectively within a reasonable time. The performance is then evaluated by turning the model into a practical application for a fog forecasting problem. Fog forecasting is an imbalanced training sets problem, since the probability of fog appearance in the observation location is very low. Fog events should be predicted every 30 minutes based on the observation of meteorological conditions. Our experiments showed that CombNET-II could achieve a high prediction rate compared to the k-nearest neighbor classifier and the three-layered multilayer perceptron trained with BP. Part of this research was presented in the 1999 Fog Forecasting Contest sponsored by Neurocomputing Technical Group of IEICE, Japan, and CombNET-II achieved the highest accuracy among the participants.

A study on improving the performance of the real-coded genetic algorithm for electromagnetic inverse scattering of two-dimensional perfectly conducting cylinders is presented. Three schemes, namely, the penalty function approach, the closed cubic B-splines local shape function approach and the adaptive hybrid algorithm approach are proposed to deal with the problem. These schemes can be used separately or be combined to improve the performance. Numerical examples validate the schemes.

We examine a concatenated code which consists of a rate 1/2, 4-state turbo code (an inner code) and a single-parity-check product code (an outer code), and discuss the decoding structure called a double concatenated decoding scheme. From our Monte Carlo simulation trials, we show the advantage of the concatenated codes over turbo codes alone. Specifically, when we use an interleaver of 4096 bits, the Eb/No to obtain a BER of 10-6 is about 1.45 dB for the concatenated code. On the other hand, it is more than 2.5 dB for the turbo code alone. So, the Eb/No improvement can be achieved by about 1 dB. This improvement in Eb/No was also obtained for the interleavers of 8192 and 2048 bits. Therefore, the concatenated codes using a double concatenated decoding scheme can solve the problem of the BER flattening in decoding of turbo codes.

This paper attempts to establish a theory for a general auto-associative memory model. We start by defining a new concept called supporting function to replace the concept of energy function. As known, the energy function relies on the assumption of symmetric interconnection weights, which is used in the conventional Hopfield auto-associative memory, but not evidenced in any biological memories. We then formulate the information retrieving process as a dynamic system by making use of the supporting function and derive the attraction or asymptotic stability condition and the condition for convergence of an arbitrary state to a desired state. The latter represents a key condition for associative memory to have a capability of learning from variant samples. Finally, we develop an algorithm to learn the asymptotic stability condition and an algorithm to train the system to recover desired states from their variant samples. The latter called sample learning algorithm is the first of its kind ever been discovered for associative memories. Both recalling and learning processes are of finite convergence, a must-have feature for associative memories by analogy to normal human memory. The effectiveness of the recalling and learning algorithms is experimentally demonstrated.

We propose a new effective method of managing flash memory space for flash memory-specific file systems based on a log-structured file system. Flash memory has attractive features such as non-volatility and fast I/O speed, but it also suffers from inability to update in situ and from limited usage (erase) cycles. These drawbacks necessitate a number of changes to conventional storage (file) management techniques. Our focus is on lowering cleaning cost and evenly utilizing flash memory cells while maintaining a balance between these two often-conflicting goals. The proposed cleaning method performs well especially when storage utilization and the degree of locality are high. The cleaning efficiency is enhanced by dynamically separating cold data and non-cold data, which is called 'collection operation.' The second goal, that of cycle-leveling, is achieved to the degree that the maximum difference between erase cycles is below the error range of the hardware. Experimental results show that the proposed technique provides sufficient performance for reliable flash storage systems.

An LMS adaptive digital filter using distributed arithmetic (DA-ADF) has been proposed. Cowan and others proposed the DA adaptive algorithm with offset binary coding for the simple derivation of an algorithm and the use of an odd-symmetry property of adaptive function space (AFS). However, we indicated that a convergence speed of this DA adaptive algorithm degraded extremely by our computer simulations. To overcome these problems, we have proposed the DA adaptive algorithm generalized with two's complement representation and effective architectures. Our DA-ADF has performances of a high speed, small output latency, a good convergence speed, small-scale hardware and lower power dissipation for higher order, simultaneously. In this paper, we analyze a convergence condition of DA adaptive algorithm that has never been considered theoretically. From this analysis, we indicate that the convergence speed is depended on a distribution of eigenvalues of an auto-correlation matrix of an extended input signal vector . Furthermore, we obtain the eigenvalues theoretically. As a result, we clearly show that our DA-ADF has an advantage of the conventional DA-ADF in the convergence speed.

In optical networks, wavelength converters are required to improve the efficiency of wavelength-division multiplexing. In this paper, we propose a genetic algorithm to determine the optimal locations of the nodes in the network where a given number of converters are placed. Optimality is achieved by the minimum wavelength blocking probability. Our algorithm is applied to two realistic networks constructed from the locations of major cities in Ibaraki Prefecture and from those in Kanto District in Japan and is shown to reach the nearly optimal solution in a limited number of generations. The accuracy is verified by simulation. The computational time is compared with that of an exhaustive search algorithm.

A kind of online edge-coloring problems on bipartite graphs is considered. The input is a graph (typically with no edges) and a sequence of operations (edge addition and edge deletion) under the restriction that at any time the graph is bipartite and degree-bounded by k, where k is a prescribed integer. At the time of edge addition, the added edge can be irrevocably assigned a color or be left uncolored. No other coloring or color alteration is allowed. The problem is to assign colors as many times as possible using k colors. Two algorithms are presented: one with competitiveness coefficient 1/4 against oblivious adversaries, and one with competitiveness coefficient between 1/4 and 1/2 with the cost of requiring more random bits than the former algorithm, also against oblivious adversaries.

In order for a service discipline to be used for guaranteed service networks at very high speed, its overall implementation must be scalable while it provides as wide a network schedulability region as possible. From this point of view, GPS-based service disciplines provide a narrow network schedulability region while EDF-based disciplines suffer from the implementation complexities of rate-controllers and admission control. Alternatively, although service disciplines based on service-curves can provide a wider network schedulability region than GPS-based and EDF-based disciplines, they may have even worse implementation complexities than EDF-based disciplines. In this paper, we propose to employ a service discipline based on our specific service-curves. We show that our service discipline has comparable implementation complexity to GPS-based disciplines while providing the same wide network schedulability region that EDF-based disciplines can provide. In fact, this service discipline is an SCED service discipline proposed in [14]. However, our specific service-curves provide the SCED service discipline with the same network schedulability region that EDF-based disciplines can provide, O(1) complexity for deadline calculation, and O(N) complexity for admission control where N is the number of sessions.

We propose two novel blind LMS algorithms, called exponential step size LMS algorithms (ES-LMS), for adaptive array antennas with fast convergence speeds. Both of the proposed algorithms are much better at tracking signal sources than the conventional LMS algorithms. In addition, they require neither spatial knowledge nor reference signals since they use the finite symbol property of digital signal. Computer simulations verify performances of the two proposed algorithms.

In this paper, we present a method to analyze the steady-state performance of a complex coefficient adaptive IIR notch filter which is useful for the rejection of multiple narrow-band interferences from broad-band signals in quadrature phase shift keying (QPSK) spread-spectrum communication systems. The adaptive notch filter based on the simplified gradient algorithm is considered. Analytical expressions have been developed for the conditional mean and variance of notch filter output. The signal-to-noise ratio improvement factor is also obtained from which the validity of the use of the notch filter can be concluded. Finally, the results of computer simulations are shown which confirm the theoretical predictions.