We propose the kernel-based pattern recognition hardware and its design methodology using the genetic algorithm. In the proposed design methodology, pattern data are transformed into the truth tables and the truth tables are evolved to represent kernels in the discrimination functions for pattern recognition. The evolved truth tables are then synthesized to logic circuits. Because of this data direct implementation approach, no floating point numerical circuits are required and the intrinsic parallelism in the pattern data set is embedded into the circuits. Consequently, high speed recognition systems can be realized with acceptable small circuit size. We have applied this methodology to the image recognition and the sonar spectrum recognition tasks, and implemented them onto the newly developed FPGA-based reconfigurable pattern recognition board. The developed system demonstrates higher recognition accuracy and much faster processing speed than the conventional approaches.

This paper presents a new method that efficiently generates all of the kernels of a sum-of-products expression. Its main feature is the memorization of the kernel generation process by using a graph structure and implicit cube set representations. We also show its applications for common logic extraction. Our extraction method produces smaller circuits through several extensions than the extraction method based on two-cube divisors known as best ever.

Excitation of magnetostatic surface waves by slot line transducers is analyzed by using the integral kernel expansion method. The Fourier integral for the current density is derived in terms of an unknown normal component of the magnetic flux density in a slot region. The integral kernel is expanded into a series of orthogonal polynomials and then applying Galerkin's method to the resulting equation yields a system of linear equations for the unknown coefficients. Comparison of a numerical result by the present method with an experiment is in good agreement.

In this paper we present a distributed routing protocol for finding two node-disjoint paths between each pair of nodes in a computer network. In the proposed protocol, each node in the network has the same procedure, which is driven by local information with respect to the network topology, such as adjacent nodes on a spanning tree in the network. Thus, the execution of the protocol can continue after changes of the network topology and load. Then, a spanning tree-based kernel construction is introduced to synchronize procedures under the distributed control of the protocol. Additionally, the routing scheme based on the protocol possesses the enhanced capabilities of alternate routes and load splitting, which cope with failures and load variations in the network. Thus, even if topology changes damage the obtained disjoint paths, the paths themselves can be updated efficiently.

Excitation of magnetostatic surface waves by coplanar waveguide transducers is analyzed by using the integral kernel expansion method. The Fourier integral for the current density is derived in terms of an unknown normal component of the magnetic flux density on slot region of a coplanar waveguide. The integral kernel is expanded into a series of Legendre polynomials and then applying Galerkin's method to the unknown field reduces the Fourier integral to a system of linear equations for the unknown coefficients. In this process, we should take into account the edge conditions which show nonreciprocal characteristics depending on frequency. The present method shows excellent agreement with experiments.

A factorization, which provides a factored form, is an extremely important part of multi-level logic synthesis. The number of literals in a factored form is a good estimate of the complexity of a logic function, and can be translated directly into the number of transistors required for implementation. Factored forms are described as either algebraic or Boolean, according to the trade-off between run-time and optimization. A Boolean factored form contains fewer number of literals than an algebraic factored form. In this paper, we present a new method for a Boolean factorization. The key idea is to build an extended Boolean matrix using cokernel/kernel pairs and kernel/kernel pairs together. The extended Boolean matrix makes it possible to yield a Boolean factored form. We also propose a heuristic method for covering of the extended Boolean matrix. Experimental results on various benchmark circuits show the improvements in literal counts over the algebraic factorization based on Brayton's Boolean matrix.

In this paper, an algorithm for Boolean factoring is presented. The algorithm is based on a technique of rectangle covering. A distinctive feature of the algorithm is that no minimization step is required to achieve Boolean factoring. The method for computing Boolean products rests on the concepts of super-product, extended kernel and extended co-kernel-cube matrix. Results of a comparison with the algorithms GOOD_FACTOR and QUICK_FACTOR implemented in SIS are presented. SIS is a program for logic synthesis developed at the University of Berkeley. All performed tests indicate that the proposed algorithm realizes a good tradeoff between factoring quality and computing time.

In order to investigate the nonlinearity and color responses of visual evoked potentials (VEPs), which have been useful in objectively detecting human color vision characteristics, a nonlinear system identification method was applied to VEPs elicited by isoluminant color stimuli, and the relationship between color stimuli and VEPs was examined. VEPs of normal subjects elicited by chromatically modulated stimuli were measured, and their binary kernels were estimated. Results showed that a system with chromatically modulated stimuli and VEP responses can be expressed by binary kernels up to the second order and that first- and second-order binary kernels depended on the color of the stimulus. The characteristics of second-order kernels reflected the difference between two chromatic channels. Opponent-color responses were included in first-order binary kernels, suggesting that they could be used as an index to test human color vision.

The integral kernel expansion method is applied to an analysis of scattering of magnetostatic forward volume waves (MSFVWs) by an array with any number of metal strips. In this method, first the integral kernel of the Fourier integral is expanded in terms of orthogonal polynomials to obtain moment equations. Then a system of algebraic equations is derived by applying the Galerkin's method. In the process, interaction between strips is naturally taken into account and real current distributions on the strips are determined such that boundary conditions are satisfied. Calculus confirmation through the energy conservation principle shows that numerical results are quite satisfactory. A comparison shows that theoretical results are in good agreement with experimental ones except the vicinity of lower and upper limits of the MSFVW band. It is shown that an infinite number of propagation modes is excited even if a wave of single mode is incident. Dependence of the scattering on dimension of arrays and on frequency and mode of an incident wave is obtained.

We propose the Kernel MUSIC algorithm as an improvement over the conventional MUSIC algorithm. This algorithm is based on the orthogonality between the image and kernel space of an Hermitian mapping constructed from the received data. Spatial smoothing, needed to apply the MUSIC algorithm to coherent signals, is interpreted as constructing procedure of the Hermitian mapping into the subspace spanned by the constituent vectors of the received data. We also propose a new spatial smoothing technique which can remove the redundancy included in the image space of the mapping and discuss that the removal of redundancy is essential for improvement of resolution. By computer simulation, we show advantages of the Kernel MUSIC algorithm over the conventional one, that is, the reduction of processing time and improvement of resolution. Finally, we apply the Kernel MUSIC algorithm to the Laser Microvision, an optical misroscope we are developing, and verify that this algorithm has about two times higher resolution than that of the Fourier transform method.

It is an important problem in fields of radar, sonar, and so on to estimate parameters of closely spaced multiple signals. The MUSIC algorithm with the forward-backward (FB) spatial smoothing is considered as the most effective technique at present for the problem with coherent signals in a variety of fields. We have applied this in Laser Microvision. Recently, Shimotahira has proposed the Kernel MUSIC algorithm, which is applicable to cases when signal vectors and noise vectors are orthogonal. It also utilizes Gaussian elimination of the covariance matrix instead of eigenvalue analysis to estimate noise vectors. Although the amount of computation by the Kernel MUSIC algorithm became lighter than that of the conventional MUSIC algorithm, the covariance matrix was formed to estimate noise vectors and also all noise vectors were used to analyze the MUSIC eigenspectrum. The heaviest amount of computation in the Kernel MUSIC algorithm exists in the transformation of the covariance matrix and the analysis of the MUSIC eigenspectrum. We propose a more straightforward algorithm to estimate noise vectors without forming a covariance matrix, easier algorithm to analyze the MUSIC eigenspectrum. The superior characteristics will be demonstrated by results of numerical simulation.

In the usual optical flow detection, the gradient constraint, which expresses the relationship between the gradient of the image intensity and its motion, is combined with the least-squares criterion. This criterion means assuming that only the time derivative of the image intensity contains noise. In this paper, we assume that all image derivatives contain noise and derive a new optical flow detection technique. Since this method requires the knowledge about the covariance matrix of the noise, we also discuss a method for its estimation. Our experiments show that the proposed method can compute optical flow more accurately than the conventional method.

Time-frequency representations (TFRs) have been developed as tools for analysis of non-stationary signals. Signal dependent TFRs are known to perform well for a much wider range of signals than any fixed (signal independent) TFR. This paper describes customised and sequential versions of the signal dependent TFR proposed in [1]. The method, which is based on the use of the Radon transform at distance zero in the ambiguity domain, is simple and effective in dealing with both simulated and real data. The use of the described method for time-scale analysis is also presented. In addition, the paper investigates a simple technique for detection of noisy chirp signals using the Radon transfrom in the ambiguity domain.

In this paper, we propose a method, called Kernel Hidden Unit Analysis, to reduce the network size. The kernel hidden unit analysis in composed of two principal components: T-component and S-component. The T-component transforms original networks into the networks which can easily be simplified. The S-component is used to select kernel units in the networks and construct kernel networks with kernel units. For the T-component, an entropy function is used, which is defined with respect to the state of the hidden units. In a process of entropy minimization, multiple strongly inhibitory connections are to be generated, which tend to turn off as many units as possible. Thus, some major hidden units can easily be extracted. Concerning the S-component, we use the relevance and the variance of input-hidden connections and detect the kernel hidden units for constructing the kernel network. Applying the kernel hidden unit analysis to the symmetry problem and autoencoders, we perfectly succeeded in obtaining kernel networks with small entropy, that is, small number of hidden units.

The emerging discipline of responsive systems demands fault-tolerant and real-time performance in uniprocessor, parallel, and distributed computing environments. The new proposal for responsiveness measure is presented, followed by an introduction of a model for responsive computing. The model, called CONCORDS (CONsensus/COmputation for Responsive Distributed Systems), is based on the integration of various forms of consensus and computation (progress or recovery). The consensus tasks include clock synchronization, diagnosis, checkpointing scheduling and resource allocation.

The mechanisms for executing concurrent applications proposed so far fall into one of three groups: processes, kernel-level threads, and user-level threads. Each of them is insufficient in terms of either parallelism, the flexibility to combine separately developed programs at run-time, or costs of operations such as creation, switching, and termination. A thread facility in the XERO operating system overcomes this problem and provides a uniform framework for executing concurrent applications. To achieve parallelism of threads, the flexibility to combine separately developed programs at run-time, and fast thread operations, the operating system kernel and a thread management module in a user address space manage threads cooperatively. We implemented the cooperative thread management mechanism and measured its performance to examine the effectiveness of our approach.