摘要:

AbstractThe feature of the parallel computer, where the tasks are allocated to the processors at the compiling stage, is that the control at the execution stage is simplified. Thus, a high‐speed execution is expected. However, a problem may arise when the processing of each task cannot accurately be estimated before the execution due to the uncertainty such as the delay in the network or the cache‐miss, making the performance deterioration inevitable.In the allocation of more than one executable task to the idle processors in the conventional scheduling algorithm, it is not particulary specified which task is allocated to which processor. This paper notes the forementioned point, and attempts to suppress the performance deterioration to the minimum by deciding the foregoing point by an adequate algorithm. More precisely, the basic idea is to reduce the precedence constraint by allocating two tasks with precedence constraint to the same processor.A new scheduling algorithm is proposed and is evaluated using a number of task graphs. It is shown that the speed improvement of 4 to 16 percent can be achieved, compared to the conventional algori

ISSN:0882-1666    

DOI:10.1002/scj.4690240801

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractA reconfiguration scheme is proposed in which a mesh‐connected highly parallel computer is divided into groups of PEs with small mesh‐structures, a spare row is added to each group (in what follows, such a group with a spare row is called a plane), these planes are successively connected upward and downward, and finally the top and bottom groups are connected. The scheme has such features that: (1) although switchings for reconfiguration are done locally, compensations are done globally, considering the distribution of faults over the whole planes; and (2) switching algorithm and circuits are simple and hence our scheme is suitable for dynamic reconfiguration.First, a method for repairing faults is described, and the necessary and sufficient condition for repairability is given. Next, formulas for the reliabilities of systems are given. Using these formulas, an example of computing the improvement degree of MTTF is illustrated and the result is compared with those in the literature. The probabilities of system survivals against the number of faulty PE's also are analyzed and the results are compared with those in the literature. Finally, logic circuits for the reconfiguration are shown and the correctness of their behavior is pro

ISSN:0882-1666    

DOI:10.1002/scj.4690240802

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractIn most of the past theoretical studies of the parallel processing of the loop, the discussion developed assuming that all execution times of the iterations are the same. Especially, the discussion is always developed under such an assumption for the doacross loop, which is the general loop processing model, including the doall and doserial loops. Considering the multiprocessor memory architectures at present and in the future, however, the following assumption seems more adequate.In each of the processors participating in the loop processing, the execution times of the second and the following iterations are shorter than that of the first. Under this assumption, there exists the number of processors (called the optimal number of processors) that minimizes the loop processing time.This paper considers the single doacross loop, and discusses the following loop processing: (1) after the delay time ofdfrom the start of execution of thei‐th iterationIi, the next iterationIi+1is permitted; (2) the iteration that once started its execution is never preempted until the execution is completed; (3) for each processor participating in the loop processing, let the execution time for the first iteration beB, and that of the second and the following iterations beb.Then,d

ISSN:0882-1666    

DOI:10.1002/scj.4690240803

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractA library program is developed which makes it easy for users to observe a whole tree and some specified parts of the tree simultaneously, and also facilitates the input of commands for the tree by allowing direct manipulation on the display. This capability should be useful in a variety of application programs. By measuring the update time of large tree structures, the calculation of the layout algorithm, and the update time of the display, it is shown that VTM is usable in practice for large trees.

ISSN:0882-1666    

DOI:10.1002/scj.4690240804

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractThis paper is based on the concept of the learning function, which represents the input‐output relation of the learning algorithm. The learning process and the information compression process are formulated as the PAC learning function and the Occam function, respectively, and their equivalence is discussed. It is shown that the Occam function is always a consistent PAC learning function, while its converse is not always true.The weak Occam function which is obtained by weakening the condition concerning the information compression power of Occam function is defined anew and it is shown that the weak Occam function is always a consistent PAC learning function. Furthermore, a procedure is shown which derives the weak Occam function from the PAC learning function under a certain conditio

ISSN:0882-1666    

DOI:10.1002/scj.4690240805

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractThis paper proposes a method for tracking an object contour in a sequence of images. In this method, a contour model is defined by elastics. The contour tracking problem is formulated as an optimization problem to find the position that minimizes both the elastic energy of its model and the potential energy derived from the edge potential image that includes a target object contour. The contour model is updated sequentially according to the tracking result in the previous frame.In the proposed approach, the contour model not only moves toward the object contour in the current frame, but also preserves the shape of the tracked contour in the previous frame. Therefore, the proposed method makes it possible to establish object tracking even when complex texture and occluding edges exist in or near the target object. Moreover, in this method, since smooth potential fields extending over a long distance are generated by distance transformation, unlike conventional methods, long‐range motion tracking can be performed.An algorithm is newly presented which efficiently solves energy minimization problems. The validity and usefulness of the proposed method is shown with some experimental result

ISSN:0882-1666    

DOI:10.1002/scj.4690240806

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractThis paper presents a speech synthesis system using the neural network. While there have been a large number of speech synthesis systems developed, most of them are rule‐based systems for a restricted range of languages, describing the expert knowledges for the language in the form of rules. Thus problems arise in that a large amount of effort is required in developing the system, and the extension of other languages is not easy.In the area of natural language processing, on the other hand, there are practical applications of machine translation, and it is desired that speech synthesis systems for various languages be developed in the future. From such a viewpoint, the authors developed the English speech synthesis system, aiming at the easy extension to multiple languages without requiring expert knowledge.In this paper, the neural networks is applied to the part‐of‐speech detection function, the syntactic boundary detection function and the letter‐to‐phoneme conversion function, all of which have stronger language dependencies among the components of the speech synthes

ISSN:0882-1666    

DOI:10.1002/scj.4690240807

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractThis paper proposes a new training method for the phoneme identification neural network called “neural fuzzy training.” In the proposed training, nondeterministic (fuzzy) class information is assigned to the training signal, in contrast to the traditional method where a deterministic class information is assigned.This study aims at the realization of a robust neural network, thereby improving the cumulative recognition rate of the phoneme identification and avoiding overtraining. The proposed neural fuzzy training is realized by backpropagation. In the conventional training, a deterministic phoneme class information is assigned to the training signal of the neural network as the value 1 or 0. However, in the proposed training, the fuzzy class information is assigned to the training signal for each training sample as the likelihood value between 0 and 1.In the proposed training method, the likelihood is calculated by the monotonically decreasing function (such as exp(−α ·d2)) of the distance between the training sample and the closest sample belonging to each phoneme class. The proposed neural fuzzy training method has a problem in that a large amount of computation cost is required since the training signal is determined by calculating the distances to all training samples. To solve this problem, the representative samples in each phoneme class are defined and the likelihood to the phoneme classes are determined by calculating the distance between the representative sample and the training sample.By this simplification of the likelihood calculation, the computational cost to determine the training signal is reduced considerably. To demonstrate the usefulness of the neural fuzzy training, an experiment is conducted: /b, d, g, m, n, N/ identification, 18 consonant identification and phrase recognition using TDNN‐LR. The ATR database is used in the experiment. In the phoneme identification experiment, the speech samples which are extracted using the hand‐label is used. The TDNN is trained using speed samples uttered in word style, and the evaluation is performed using speech samples uttered in phrase style and in sentence style.In the phrase recognition experiment using TDNN‐LR, the TDNN is trained using speed samples uttered word style using a hand label. The evaluation is performed using speech samples uttered in phrase style. In either experiment, an improvement of using the fuzzy training can be observed. Especially, in the phrase recognition experiment using TDNN‐LR, the top recognition rate is improved from 71.2 percent to 80.9 percent, and the top 5th recognition rate is improved from 92.8 percent to 96.O percent. Furthermore, it appeared also that the neural fuzzy training is a high‐spee

ISSN:0882-1666    

DOI:10.1002/scj.4690240808

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY

摘要:

AbstractMethods of inputting stereoscopic information to neural networks and of imaging by ultrasonic sensors were examined from the perspective of three‐dimensional object recognition, and a new recognition method was developed for depth images in neural networks. Because a perceptron neural network is used as the recognition stage and ultrasonic sensors arranged in an array for both reception and transmission are used as the shape measurement stage, this method performs compression of the stereoscopic information of an object into a frame of depth information by measuring the planar distribution of the distances to the object's surface (using a pulse echo method); it has the advantage of being able to efficiently recognize an object freely moving in any direction (forward and backward, right and left, and up and down) and having any orientation within a plane, by using depth information. Since a neural network is used, even when different features are extracted from the object, direct recognition is possible. In this paper, which addresses the issue of three‐dimensional object recognition using perceptrons, we describe a measurement and recognition system capable of preprocessing stereoscopic information that can be freely positioned in three dimensions to allow recognition by perceptrons, and we present the results of recognition tests. Finally, we propose a recognition algorithm that actively uses depth (which is not a feature in two dimensions), show that the algorithm can recognize both the position and orientation of the object, and demonstrate its practical

ISSN:0882-1666    

DOI:10.1002/scj.4690240809

出版商:Wiley Subscription Services, Inc., A Wiley Company    

年代:1993

数据来源: WILEY