The Hopfield neural network for optimization problems often falls into local minima. To escape from the local minima, the neuron unit in the neural network is modified to become an oscillatory unit by adding a simple self-feedback circuit. By combining the oscillatory unit with an energy-value extraction circuit, an oscillatory neural network is constructed. The network can repeatedly extract solutions, and can simultaneously evaluate them. In this paper, the network is applied to four NP-complete problems to demonstrate its generality and efficiency. The network can solve each problem and can obtain better solutions than the original Hopfield neural network and simple algorithms.

The fault tolerant computer (FTC) is applied as a communication or database server in the information service and computer aided process control fields. User requires of the FTC are to provide the current level of performance and software transparency needing no additional dedicated program for fault tolerance. To meet these requirements, we propose quadprocessor redundancy (QPR) architecture that combines dualRISC based duplicated CPUs integrating main memories, and duplicated I/O subsystems by using some additional hardware. Duplicated CPUs run under the instruction level synchronization (lock step operation) , and the duplicated I/O subsystems are managed by an operating system. When a fault is detected, the faulty CPU is isolated by hardware. User program is continuously executed by the remaining CPU. We applied the QPR to our UNIX servers, and achieved satisfactory levels of performance.

So far, neuronal dendrites have been characterized as electrically passive cables. However, recent physiological findings have revealed complex dynamics due to active conductances distributed over dendrites. In particular, the voltage-gated calcium and calcium-activated conductances are essential for producing diverse neuronal dynamics and synaptic plasticity. In this paper, we investigate the functional significance of the dendritic calcium-activated dynamics by computer simulations. First, the dendritic calcium-activated responses are modeled in a discrete compartmental form based on the physiological findings. Second, the basic stimulus-response characteristics of the single compartment dendrite model are investigated. The model is shown to reproduce the neuronal responses qualitatively. Third, the spatio-temporal dynamics of the dendrite shafts are modeled by longitudinally connecting 10 single compartments with coupling constants which are responsible for the dendrite thickness. The thick dendrite models, corresponding to proximal dendrites, respond in a spatially cooperative manner to a localized constant or periodic current stimulation. In contrast, the highly activated compartments are forced to be localized in the neighborhood of the stimulation-site in the fine dendrite models corresponding to distal dendrites. These results suggest that dendritic activities are spatially cooperated in a site-dependent manner.

This paper investigates the accepting powers of multi-inkdot two-way alternating pushdown automata (Turing machines) with sublogarithmic space and constant leaf-size. For each k1, and each m0, let weak-ASPACEm [L(n),k] denote the class of languages accepted by simultaneously weakly L(n) space-bounded and k leaf-bounded m-inkdot two-way alternating Turing machines, and let strong-2APDAm[L(n),k] denote the class of languages accepted by simultaneously strongly L(n) space-bounded and k leaf-bounded m-inkdot two-way alternating pushdown automata. We show that(1) strong-2APDAm [log log n,k+1]weak-ASPACEm[o(log n),k]φfor each k1 and each m1, and(2) strong-2APDA(m+1) [log log n,k]weak-ASPACEm[o(log n),k]φfor each k1 and each m0.

In this paper, we show that a parity function with n variables can be computed by a threshold circuit of depth O((log n)/c) and size O((2clog n)/c), for all 1c [log(n+1)]-1. From this construction, we obtain an O(log n/log log n) upper bound for the depth of polylogarithmic size threshold circuits for parity functions. By using the result of Impagliazzo, Paturi and Saks[5], we also show an Ω (log n/log log n) lower bound for the depth of the threshold circuits. This is an answer to the open question posed in [11].

A generalized Reed-Muller expression (GRM) is obtained by negating some of the literals in a positive polarity Reed-Muller expression (PPRM). There are at most 2(n2)^(n-1) different GRMs for an n-variable function. A minimum GRM is one with the fewest products. This paper presents certain properties and an exact minimization algorithm for GRMs. The minimization algorithm uses binary decision diagrams. Up to five variables, all the representative functions of NP-equivalence classes were generated and minimized. Tables compare the number of products necessary to represent four-and five-variable functions for four classes of expressions: PPRMs, FPRMs, GRMs and SOPs. GRMs require, on the average, fewer products than sum-of-products expressions (SOPs), and have easily testable realizations.

Mechanism of the noise generation caused by the optical feedback in semiconductor laser was experimentally determined. Two types of the mode competition phenomena were confirmed to be the generating mechanisms. Applicability of the self-sustained pulsation to be a noise reduction method was also discussed.

In [2] Ibarra introduced a restricted version of one-way multihead pushdown automaton (PDA), called a simple one-way multihead PDA, and showed that such machines recognize only languages with semilinear property. The main result of this paper is that for each k 1, simple (sensing) one-way (k + 1)-head PDA's are more powerful than simple (sensing) one-way k-head PDA's. This paper also investigates closure properties for simple (sensing) one-way multihead PDA's

In this paper, we consider the protein structure alignment problem, which is a very important problem in molecular biology. Since an outline of protein structure is represented by a sequence of points in three-dimensional space, this problem is defined as the following geometric pattern matching problem: given two point sequences P and Q in three-dimensions and a real number δ > 0, find a maximum-cardinality set of point pairs such that the distance between each pair is at most δ under the condition that any translation and rotation can be applied to P. Since it is very difficult to solve this problem exactly, we consider algorithms that solve it approximately. We propose three algorithms: BASICALIGN, RANDALIGN and FRAGALIGN whose worst case time complexities are O(n8), O((n7/k3) polylog(n)) and O(n4) respectively, where n denotes the size of larger input structure and k denotes the minimum size of the alignment to be obtained. All of these have the following common framework: a series of initial superpositions are computed; for each of such superpositions, a rough alignment is first computed using a dynamic programming technique, and then it is refined through an iterative improvement procedure which also uses dynamic programming; the best alignment among them is selected as an output. The difference among three algorithms lies in the methods of finding initial superpositions. BASICALIGN, RANDALIGN and FRAGALIGN use exhaustive search, random sampling technique and fragment-based search, respectively. We prove guaranteed approximation ratios (in the sense of distances between point pairs) for theoretical versions of BASICALIGN and RANDALIGN. Practical versions of RANDALIGN and FRAGALIGN were implemented and compared with a previous algorithm using real protein structure data. The experimental results show that FRAGALIGN is best among them and it outputs good alignments quickly.

For reversal complexity on an off-line Turing machine, which is a Turing machine with a read-only two-way input tape except work-tapes, we can consider two kinds of definition; the first one is a definition in which the number of reversals over the input tape is not counted, and the second one is a definition in which it is counted. Unlike time and space complexities, whether or not there is any difference between these two definitions does not seem to be trivial. In this paper, we will show the following results: (1) let S(n) be any function, and R(n) be an (R(n), S(n)) reversal-space constructible function. Then, DRESPk(R(n), S(n)) IDRESPk+2(R(n) + log(nS(n)), n2R(n)S(n)), (2) let R(n) and S(n) be any functions. Then, NRESPk(R(n), S(n)) INRESPk+1(R(n), n2S(n)), and ARESPk(R(n), S(n)) = IARESPk(R(n), S(n)), where DRESP denotes a deterministic reversal- and space-bounded class under the definition disregarding reversals over the input tape, and IDRESP denotes a deterministic reversal- and space-bounded class under the definition counting it. The suffix k denotes the number of work-tapes. The classes NRESP, INRESP, ARESP and IARESP are also defined similarly for NTMs and ATMs.

This paper proposes a method for deriving protocol specifications in a communicating processes model, in which protocol specifications are modeled by finite automata communicating through the LOTOS multirendezvous mechanism. Message sequence charts (MSCs) are used for the derivation inputs. MSCs are graphical representations of traces of protocols and are suitable for defining requirements. Since an MSC usually covers only partial behavior, several mechanisms for composing a large number of MSCs from element MSCs have been proposed. These mechanisms, however, are not adequate: Either their input language for MSCs is not powerful enough, or they need some information on protocol specifications (i.e., implementation specifications). This paper proposes the use of regular expressions over MSCs to fully define protocols. The proposed language is powerful enough to describe protocols in the communicating processes model. A derivation algorithm based on a finite automata construction algorithm that accept sets expressed by regular expressions is presented. Because the derived protocols sometimes include unrequired behavior, an algorithm for detecting unrequired behavior is also presented.

In this paper, we propose a camera calibration method that estimates both intrinsic parameters (perspective and distortion) and extrinsic parameters (rotational and translational). All camera parameters can be determined from one or more images of planar pattern consists of parallelogramatic grid points. As far as the pattern can be visible, the relative relations between camera and patterns are arbitrary. So, we have only to prepare a pattern, and take one or more images changing the relative relation between camera and the pattern, arbitrarily; neither solid object of ground truth nor precise z-stage are required. Moreover, constraint conditions that are imposed on rotational parameters are explicitly satisfied; no intermediate parameter that connected several actual camera parameters are used. Taking account of the conflicting fact that the amount of distortion is small in the neighborhood of the image center, and that small image has poor clues of 3-D information, we adopt iterative procedure. The best parameters are searched changing the size and number of parallelograms selected from grid points. The procedure of the iteration is as follows: The perspective parameters are estimated from the shape of parallelogram by nonlinear optimizations. The rotational parameters are calculated from the shape of parallelogram. The translational parameters are estimated from the size of parallelogram by least squares method. Then, the distortion parameters are estimated using all grid points by least squares method. The computer simulation demonstrates the efficiency of the proposed method. And the results of the implementation using real images are also shown.

It is well known that extracting parallel loops plays a significant role in designing parallelizing compilers. The execution efficiency of a loop is enhanced when the loop can be executed in parallel or partial parallel, like a DOALL or DOACROSS loop. This paper reports on the practical parallelism detector (PPD) that is implemented in PFPC (a portable FORTRAN parallelizing compiler running on OSF/1) at NCTU to concentrate on finding the parallelism available in loops. The PPD can extract the potential DOALL and DOACROSS loops in a program by invoking a combination of the ZIV test and the I test for verifying array subscripts. Furthermore, if DOACROSS loops are available, an optimization of synchronization statement is made. Experimental results show that PPD is more reliable and accurate than previous approaches.

A siphon (or alternatively a structutal deadlock) of a Petri net is defined as a set S of places such that existence of any edge from a transition t to a place of S implies that there is an edge from some place of S to t. A minimal siphon is a siphon such that any proper subset is not a siphon. The results of the paper are as follows. (1) The problem of deciding whether or not a given Petri net has a minimum siphon (i.e., a minimum-cardinality minimal siphon) is NP-complete. (2) A polynomial-time algorithm to find, if any, a minimal siphon or even a maximal calss of mutually disjoint minimal siphons of a general Petri net is proposed.

This paper suggests modified LZSS which is suitable for compressing Hangul data by Hangul character token and the string token with small size based on Hangul properties. The Hangul properties can be described in 2 ways. 1) The structure of a Hangul character consists of 3 letters: The first sound letter, the middle sound letter, and the last sound letter which are called Cho-seong, Jung-seong, and Jong-seong, respectively. 2) The code of Hangul is represented by 2 bytes. The first property is used for making the character token processing Hangul characters which occupies most of the unmatched characters. That is, the unmatched Hangul characters are replaced with one Hangul character token represented by Huffman codes of Cho-seong, Jung-seong, and Jong-seong in regular sequence, instead of 2 character tokens. The second property is used to shorten the size of the string token processing matched string. In other words, since more than 75% of Hangul data are Hangul and Hangul codes are constructed in 2 bytes, the addresses of the window of LZSS can be assigned in 2-byte unit. As a result, the distance field and the length field of the string token can be lessened by one bit each. After compressing Hangul data through these tokens, about 3% of improvement could be made in compression ratio.

For reduction of computational complexity in the IA algorithm, the thinned-out IA algorithm in which only one step size is updated every iteration is proposed and is complementarily switched with the HA algorithm according to the convergence. The switching is determined by using the gradient of the error signal power. These are investigated through the computer simulations.

The multichannel distortions of direct modulated laser diode were studied from the view point of rate equations. A novel technique for compensating the composite second order distortion (CSO) was proposed. Meanwhile, the related calibration procedures were indicated. After the compensation, 10 dB improvement in CSO was obtained

A minimal siphon (or alternatively a structural deadlock) of a Petri net is defined as a minimal set S of places such that existence of any edge from a transition t to a place of S implies that there is an edge from some place of S to t. The subject of the paper is to find a minimal siphon containing a given set of specified places of a general Petri net.

This paper proposes a novel Doppler frequency shift compensation technique to achieve terrestrial and low earth orbit (LEO) satellite dual mode DS/CDMA terminals robust to high Doppler shift and multipath fading. In order to satisfy the requirements of wide dynamic range and high accuracy simultaneously, the proposed scheme employs two stage compensation scheme, i.e., coarse compensation to keep dynamic range of about 100 kHz and fine compensation to satisfy its resolution of about 30 Hz, using block demodulation technique. Computer simulation results show that the proposed scheme can sufficiently compensate for the offset frequency up to the range of about 100 kHz at the terrestrial and LEO satellite combined mobile communication systems.

Analog computation is a processing method that solves problems utilizing an analogy of a physical system to the problem. As it is based on actual physical effects and not on symbolic operations, it is therefore a promising architecture for quantum processors. This paper presents an idea for relating quantum structures with analog computation. As an instance, a method is proposed for solving an NP-complete (nondeterminis-tic polynomial time complete) problem, the three-color-map problem, by using a quantum-cell circuit. The computing process is parallel and instantaneous, so making it possible to obtain the solution in a short time regardless of the size of the problem.