In an adaptive load balancing, the location policy to determine a destination node for transferring tasks can be classified into three categories: dynamic selection, random selection, and state polling. The dynamic selection immediately determines a destination node by exploiting the state information broadcasted from other nodes. It not only requires the overheads of collecting the state information, but may cause an unpredictable behavior unless the state information is accurate. Also, it may not guarantee even load distribution. The random selection determines a destination node at random. The state polling determines a destination node by polling other nodes. It may cause some problems such as useless polling, unachievable load balancing, and system instability. A new Sender-initiated Adaptive LOad balancing scheme (SALO) is presented to remedy the above problems. It determines a destination node by exploiting the predictable state knowledge and by polling the destination node. It can determine a good destination with minimal useless polling and guarantee even load distribution. Also, it has an efficient mechanism and good data structure to collect the state information simply. An analytic model is developed to compare with other well known schemes. The validity of the model is checked with an event-driven simulation. With the model and the simulation result, it is shown that SALO yields a significant improvement over other schemes, especially at high system loads.

A new yield prediction model has been developed, which can successfully describe the actual chip fabrication yield. It basically consists of modeling of particles deposited on wafer surface, considering the change in their size and spatial distribution due to the subsequent processing steps and a new concept of virtual line width in pattern layouts. It is confirmed that this yield prediction model serves as an effective navigator for improvement/optimization of fabrication lines such as pointing out the process step/equipments to be modified for yield improvements.

Caenorhabditis elegans during feeding gives good moving biological images",in which motions of several pulsing organs are superposed on its head swing. A powerful method to extract dynamic features is presented. First step is to use a variance picture VAG4 in order to pick up active pixel coordinates of concerned moving objects. Superiority of VAG4 over usual variance picture VAG2 is shown quantitatively by a model of moving particles. Pulsing areas of C. elegans, are exhibited more clearly in VAG4 than VAG2. Second step is use of a new subtraction method to extract main frequency bands. FFT spectra are averaged in active positions where VAG4 is above threshold THVR in the square with 88 pixels (ONA). The power spectra averaged in the enlarged squares (ELA) are subtracted from those in ONA, in which ELA includes ONA in its centre position. Large peak bands emerge in the subtracted power spectra. The subtraction eliminates the effect of head swing by spatial averagings in ELA. This new emphasizing method is compared to another subtraction method. The characteristic frequency of periodical moving organs coincides well with the values observed by other research groups and our visual estimation of replayed VTR images. Thus the proposed extraction method is verified to work well in double superposed motions.

In previous papers the building of hierarchical networks made up of components using fuzzy rules was presented. It was demonstrated that this approach could be used to construct networks to solve classification problems, and that in many cases these networks were computationally less expensive and performed at least as well as existing approaches based on feedforward neural networks. It has also been demonstrated how this approach could be extended to real-valued problems, such as function approximation and time series prediction. This paper investigates the problem of choosing the best network for real-valued approximation problems. Firstly, the nature of the network parameters, how they are interrelated, and how they affect the performance of the system are clarified. Then we address the problem of choosing the best values of these parameters. We present two model selection tools in this regard, the first using a simple statistical model of the network, and the second using structural information about the network components. The resulting network selection methods are demonstrated and their performance tested on several benchmark and applied problems. The conclusions look at future research issues for further improving the performance of the clustering network.

This paper presents the performance prediction method on sentence recognition system which uses a finite state word automaton. When each word is uttered separately, the relationship between word recognition score and sentence recognition score can be approximated using the number of word sequences at a minimum distance from each sentence in the task. But it is not clear that how we get this number when the finite state word automaton is used as linguistic information. Therefore, we propose the algorithm to calculate this number in polynomial time. Then we carry out the prediction using this method and the simulation to compare with the prediction on the task of Japanese text editor commands. And it is shown that our method approximates the lower limit of sentence recognition score.

This paper discusses the reliability of the Modified EBU method compared with the EBU and DSCQS methods where the small and different levels of impairments exist in the coded HDTV sequences. The subjective evaluation tests are carried out in the full and limited impairment ranges. And it is shown that the Modified EBU method is most reliable for both ranges.

Two method to predict targets which a user is about to point with a mouse on the basis of the trajectory of the mouse cursor were proposed. The effects of the interval between targets, the position of targets, the sampling interval and the number of sampling on the pointing time and the prediction accuracy were investigated. In both methods, the distance between targets had little effects on the pointing time. The prediction accuracy was found to be affected by the position of targets. In both prediction methods, the angle between the cursor movement vector and the vector which connects the current cursor position and the center of each target is calculated every st. As for Prediction Method1 that regards the target which correspond to the minimum angle continuously 5 times as the candidate target, the optimal condition of the sampling interval was found to be 0.06 sec or 0.08 sec. Concerning Prediction Method2 that calculates the angle n times and determines the minimum cumulative value as the candidate, the optimal condition of the number of sampling was 8.

The coherence in the time movement of the spectral vector sequence is modelled by a vector linear predictor. Such a model of the stop consonant transition is used for discrimination of the places of articulation of/ba/,/dha/,/da/, and/ga/. The effect of cross-channel correlation in giving improved recognition performance and also in reducing the time asymmetry of the predictive recognizer is studied. The high recognition score of vector model and the considerable differnce in the forward reverse score of the scalar model than a vecter model is highlighted in this study on a speech data of a set of four speakers.

The magnetic properties and the structure of electroless-deposited NiFeB films were investigated in comparison with those of electrodeposited NiFe films. The electroless-deposited NiFeB film with 27at% Fe content had the lowest coercivity, H, as low as 0.5 Oe with a saturation magnetic flux density, Bs, of 1.0 T. The saturation magnetostriction, λ, and the uniaxial magnetic anisotropy, Hk, were 5.010-6 and 10 Oe, respectively, which were larger than those of the conventional, electrodeposited permalloy film. The permeability of as-deposited Ni70Fe27B3 film was 1000 at 1 MHz. In order to improve the permeability, the film was heated at 200 in a magnetic field applied in the hard-axis direction to decrease the Hk value, and the permeability became 2000 at 1 MHz. The crystal structure and grain size of NiFeB and NiFe films were investigated by XRD, THEED and TEM. Both films with low Hc had an fcc structure; the grain size of the NiFeB film was smaller than 10 nm, while that of the NiFe film was larger, approximately 20 nm. The results suggested that the electroless-deposited NiFeB film had a larger magnetic anisotropy than the electrodeposited NiFe film. Moreover, the films with Hc less than 10 Oe ded not show clear difference between their TEM bright images and THEED patterns.

We present a framework for the unsupervised segmentation of time series. It applies to non-stationary signals originating from different dynamical systems which alternate in time, a phenomenon which appears in many natural systems. In our approach, predictors compete for data points of a given time series. We combine competition and evolutionary inertia to a learning rule. Under this learning rule the system evolves such that the predictors, which finally survive, unambiguously identify the underlying processes. The segmentation achieved by this method is very precise and transients are included, a fact, which makes our approach promising for future applications.

In this paper, we propose algorithm of deterministic nonlinear prediction, or a modified version of the method of analogues which was originally proposed by E.N. Lorenz (J. Atom. Sci., 26, 636-646, 1969), and apply it to the artificial time series data produced from nonlinear dynamical systems and further corrupted by superimposed observational noise. The prediction performance of the present method are investigated by calculating correlation coefficients, root mean square errors and signature errors and compared with the prediction algorithm of local linear approximation method. As a result, it is shown that the prediction performance of the proposed method are better than those of the local linear approximation especially in case that the amount of noise is large.

A deterministic pushdown automaton (dpda) having just one stack symbol is called a deterministic restricted one-counter automaton (droca). A deterministic one-counter automaton (doca) is a dpda having only one stack symbol, with the exception of a bottom-of-stack marker. The class of languages accepted by droca's which accept by final state is a proper subclass of the class of languages accepted by doca's. Valiant has proved the decidability of the equivalence problem for doca's and the undecidability of the inclusion problem for doca's. Hence the decidability of the equivalence problem for droca's is obvious. In this paper, we evaluate the upper bound of the length of the shortest input string (witness) that disproves the inclusion for a pair of real-time droca's which accept by final state, and present a new direct branching algorithm for checking the inclusion for a pair of languages accepted by these droca's. Then we show that the worst-case time complexity of our algorithm is polynomial in the size of these droca's.

This paper presents a comprehensive framework for the organization, retrieval, and adaptation of image information and meta-information in image database systems. The multi-level hierarchy of images that emphasizes the composition of visual entities (such as Human, Chair, , etc.) from its constituents (eye, leg, , etc.) is managed by a host architecture that is called the semantic tree. This architecture is shown to integrate description, numeric, and statistic image constituent's information into a compound space that is used as retrieval basis for semantic, sketch, and template image queries and several other composite query types. The core architecture based on which the semantic tree is constructed is shown to offer several new features such as simple prototyping, complex prototyping, low storage requirements, and automatic knowledge acquisition compatibility. The object oriented data model constitutes our comparison basis throughout the paper. Methods (functions) used to access image information are shown to be organized into a separate architecture called the query dictionary. This architecture is shown to offer a convenient hierarchical message passing medium using which a variety of composite queries are constructed. Interaction between semantic trees and the query dictionary is clarified through several examples. It is shown that the semantic tree architecture embraces additional networking semantic intormation through a range of relation representation models, the first of which is introduced in this paper. A new inheritance method called semantic relation spreading is introduced. Comprehensive examples are given to demonstrate the versatility of the new strategy.

This paper describes new modeling methods combining neural network and hidden Markov model applicable to modeling a time series such as speech signal. The idea assumes that the sequence is nonstationary and is a nonlinear autoregressive process whose parameters are controlled by a hidden Markov chain. One is the model where a non-linear predictor composed of a multi-layered neural network is defined at each state, another is the model where a multi-layered neural network is defined so that the path from the input layer to the output layer is divided into path-groups each of which corresponds to the state of the Markov chain. The latter is an extended model of the former. The parameter estimation methods for these models are shown, and other previously proposed models--one called Neural Prediction Model and another called Linear Predictive HMM--are shown to be special cases of the NPHMM proposed here. The experimental result affirms the justification of these proposed models.

This paper describes a method for predicting the user's next utterances in spoken dialog based on the topic transition model, named TPN. Some templates are prepared for each utterance pair pattern modeled by SR-plan. They are represented in terms of five kinds of topic-independent constituents in sentences. The topic of an utterance is predicted based on the TPN model and it instantiates the templates. The language processing unit analyzes the speech recognition result using the templates. An experiment shows that the introduction of the TPN model improves the performance of utterance recognition and it drastically reduces the search space of candidates in the input bunsetsu lattice.

This paper proposes a new pitch prediction method for 4 kbps CELP (Code Excited LPC) speech coding with 20 msec frame, for the future ITU-T 4 kbps speech coding standardization. In the conventional CELP speech coding, synthetic speech quality deteriorates rapidly at 4 kbps, especially for female and children's speech with short pitch period. The pitch prediction performance is significantly degraded for such speech. The important reason is that when the pitch period is shorter than the subframe length, the simple repetition of the past excitation signal based on the estimated lag, not the pitch prediction, is usually carried out in the adaptive codebook operation. The proposed pitch prediction method can carry out the pitch prediction without the above approximation by utilizing the current subframe excitation codevector signal, when the pitch prediction parameters are determined. To further improve the performance, a split vector synthesis and perceptually spectral weighting method, and a low-complexity perceptually harmonic and spectral weighting method have also been developed. The informal listening test result shows that the 4 kbps speech coder with 20 msec frame, utilizing all of the proposed improvements, achieves 0.2 MOS higher results than the coder without them.

This letter presents new estimation algorithm of ARMAX systems which do not always satisfy the strictly positive real (SPR) condition. We show how estimated parameters can converge to their true values based on the overparameterized system. Finally, the results of computer simulation are presented to illustrate the effectiveness of the proposed method.

The superconducting magnet on a maglev vehicle vibrate and heats up inside under the influence of various disturbances in running. We have investigated the characteristics of heating in the superconducting magnet vibrating under the electro-magnetic disturbance from the ground coils. This magnetic disturbance has a frequency component ranging widely from 0 Hz to several hundred Hz which is proportional to the speed of the maglev vehicle. It was revealed that an extreme increase of heat load on the inner vessel of the energized magnet occurred at a particular frequency and it surpassed the capacity of the refrigerator installed in the tank of the superconducting magnet. As a result of the investigation, we could identify broadly three factors of heating, and now we have good prospects of largely suppressing the heating by reducing the disturbance through the folded arrangement of the ground coils and a structural improvement of the magnet.

This paper proposes a practical training algorithm for artificial neural networks, by which both the optimally pruned model and the optimally trained parameter for the minimum prediction error can be found simultaneously. In the proposed algorithm, the conventional information criterion is modified into a differentiable function of weight parameters, and then it is minimized while being controlled back to the conventional form. Since this method has several theoretical problems, its effectiveness is examined by computer simulations and by an application to practical ultrasonic image reconstruction.

A deterministic pushdown automaton (dpda) having just one stack symbol is called a deterministic restricted one-counter automaton (droca). When it accepts by empty stack, it is called strict. A deterministic one-counter automaton (doca) is a dpda having only one stack symbol, with the exception of a bottom-of-stack marker. The class of languages accepted by strict droca's is a subclass of the class of languages accepted by doca's. Valiant has proved the decidability of the equivalence problem for doca's and the undecidability of the inclusion problem for doca's. Hence the decidablity of the equivalence problem for strict droca's is obvious. In this paper, we present a new direct branching algorithm for checking the inclusion for a pair of languages accepted by strict droca's. Then we show that the worst-case time complexity of our algorithm is polynomial with respect to these automata.