This paper proposes a new method of extracting feature attentive regions in a learnt multi-layer neural network. We difine a function which calculates the degree of dependence of an output unit on an inpur unit. The value of this function can be used to investigate whether a learnt network detects the feature regions in the training patterns. Three computer simulations are presented: (1) investigation of the basic characteristic of this function; (2) application of our method to a simpie pattern classification task; (3) application of our method to a large scale pattern classfication task.

Program transformation is a kind of optimization techniques for logic programs, which aims at transforming equally a program into an other form by exploiting some properties or information of the program, so as to make the program cheaper to evaluate. In this paper, a new kind of property of logic programs, called reducibility, is exploited in program transformation. A recursive predicate is reducible if the values of some variables in the recursive predicate are independent to the remainder part and can be detached from the predicate after finite times of expansions. After being proved that the semantic notion of reducibility can be replaced by the syntactic notion of disconnectivity of a R-graph, which is a kind of graph model to represent the behavior of formula expansions, an efficient testing and factoring algorithm is proposed. The paper also extends some existed results on compiled formulas of linear sirups, and compares with some related work.

This paper proposes the concept of information retrieval for fine arts database system on the fuzzy set theory, especially concerning to sensitive impression and location data. The authors have already reported several important formulations about the data structure and information retrieval models based on the fuzzy set theory for multimedia database. The fuzzy models of the information retrieval are implemented in the fine arts database system, which has the following features: (1) The procedure of information retrieval is formulated in the fuzzy set theory; (2) This database can treat multimedia data such as document data, sensitive impression, location information, and imagedata. (3) It is possible to retrieve the stored data based on sensitive impression and the location data such as "joyful pictures which have a mountain in the center and there is a tree in the right"; (4) Users can input impression words as a retrieval condition, and estimate their grades such as "low," "medium," and "high"; (5) For the result of information retrieval, the satisfaction grade is calculated based on fuzzy retrieval model; and (6) The stored data are about 400 fine arts paintings which are inserted by the textbook of fine arts currently used at the junior high school and high school in Japan. These features of this system give an effects of the fine arts education, and should be useful for information retrieval of fine arts. The results of this study will become increasingly important in connection with development of multimedia technology.

The currying of term rewriting systems (TRSs) is a transformation of TRSs from a functional form to an applicative form. We have already introduced an order-sorted version of currying and proved that the compatibility and confluence of order-sorted TRSs were preserved by currying. In this paper, we focus on a key property of TRSs, completeness. We first show some proofs omitted in Ref. [3]. Then, we prove that the SN (strongly normalizing) property, which corresponds to termination of a program, is preserved by currying. Finally, we prove that the completeness of compatible order-sorted TRSs is preserved by currying.

This paper describes a new approach to the scheduling problem that assigns tasks of a parallel program described as a task graph onto parallel machines. The approach handles interprocessor communication and heterogeneity, based on using both the theoretical results developed so far and a lookahead scheduling strategy. The experimental results on randomly generated task graphs demonstrate the effectiveness of this scheduling heuristic.

Reducing communication overhead is a key goal of program optimization for current scalable multiprocessors. A well-known approach to achieving this is to map tasks (indivisible units of computation) to processors so that communication and computation overlap as much as possible. In an earlier work, we developed a look-ahead scheduling heuristic for efficiently reducing communication overhead with the aim of decreasing the completion time of a given parallel program. In this paper, we report on an extension of the algorithm, which fills in the idle time slots created by interprocessor communication without increasing the algorithm's time complexity. The results of experiments emphasize the importance of optimally filling idle time slots in processors.

In this paper, we extend the Gnaedig's results on termination of order-sorted rewriting. Gnaedig required a condition for order-sorted signatures, called minimality, for the termination proof. We get rid of this restriction by introducing a transformation from a TRS with an arbitrary order-sorted signature to another TRS with a minimal signature, and proving that this transformation preserves termination.

A Wireless Sensor Network (WSN, for short) is a distributed system consisting of n sensor nodes and a base station. In this paper, we propose an energy-efficient protocol to initialize the sensor nodes in a WSN, that is, to assign a unique ID to each sensor node. We show that if an upper bound u on the number n of sensor nodes is known beforehand, for any f 1 and any small µ (0<µ<1), a WSN without collision detection capability can be initialized in O((log (1/µ) + log f)u1+µ) time slots, with probability exceeding 1-(1/f), with no sensor node being awake for more than O(log (1/µ)+ log f) time slots.

A radio network is a distributed system with no central shared resource, consisting of n stations each equipped with a radio transceiver. One of the most important parameters to evaluate protocols in the radio networks is the number of awake time slots in which each individual station sends/receives a data packet. We are interested in devising energy-efficient initialization protocols in the single-hop radio network (RN, for short) that assign unique IDs in the range [1,n] to the n stations using few awake time slots. It is known that the RN can be initialized in O(log log n) awake time slots, with high probability, if every station knows the number n of stations in the RN. Also, it has been shown that the RN can be initialized in O(log n) awake time slots even if no station knows n. However, it has been open whether the initialization can be performed in O(log log n) awake time slots when no station knows n. Our main contribution is to provide the breakthrough: we show that even if no station knows n, the RN can be initialized by our protocol that terminates, with high probability, in O(n) time slots with no station being awake for more than O(log log n) time slots. We then go on to design an initialization protocol for the k-channel RN that terminates, with high probability, in O(n/k + (log n)2) time slots with no station being awake for more than O(log log n) time slots.

Nonlinearity is an important factor in the biological neural networks. The motion perception and learning in them have been studied on the simplest type of nonlinearity, multiplication. In this paper, asymmetrical neural networks with nonlinear function, are studied in the biological neural networks. Then, the nonlinear higher-order system is discussed in the neural networks. The second-order system in the nonlinear biological system is shown to play an important role in the movement detection. From the theoretical analysis, it is shown that the third-order one does not contribute to the detection and the fourth-order one becomes to the second-order in the movement detection function. Hassenstein and Reichardt network (1956) and Barlow and Levick network (1965) of movements are similar to the asymmetrical network developed here. To make clear the difference among these asymmetrical networks, we derive α-equation of movement, which shows the detection of movement. During the movement, we also can derive the movement equation, which implies the movement direction regardless of the parameter α.

Parallelism on heterogeneous machines brings cost effectiveness, but also raises a new set of complex and challenging problems. This paper addresses the problem of estimating the minimum time taken to execute a program on a fine-grained parallel machine composed of different types of processors. In an earlier publication, we took the first step in this direction by presenting a graph-construction method which partitions a given program into several homogeneous parts and incorporates timing constraints due to heterogeneous parallelism into each part. In this paper, to make the method easier to be applied in a scheduling framework and to demonstrate its practical utility, we present an efficient implementation method and compare the results of its use to the optimal schedule lengths obtained by enumerating all possible solutions. Experimental results for several different machine models indicate that this method can be effectively used to estimate a program's minimum execution time.

This paper describes a new method to determine the 3-D position coordinates of a Lambertian surface from four shaded images acquired with an actively controlled, nearby moving point light source. The method treats both the case when the initial position of the light source is known and the case when it is unknown.

In this paper a method of recognizing waveform based on the Discrete Wavelet Transform (DWT) presented by us is applied to detecting the K-complex in human's EEG which is a slow wave overridden by fast rhythms (called as spindle). The features of K-complex are extracted in terms of three parameters: the local maxima of the wavelet transform modulus, average slope and the number of DWT coefficients in a wave. The 4th order B-spline wavelet is selected as the wavelet basis. Two channels at different resolutions are used to detect slow wave and sleep spindle contained in the K-complex. According to the principle of the minimum distance classification the classifiers are designed in order to decide the thresholds of recognition criteria. The EEG signal containing K-complexes elicited by sound stimuli is used as pattern to train the classifiers. Compared with traditional method of waveform recognition in time domain, this method has the advantage of automatically classifying duration ranks of various waves with different frequencies. Hence, it specially is suitable to recognition of signals which are the superimposition of waves with different frequencies. The experimental results of detection of K-complexes indicate that the method is effective.

The properties of the Haar Transform (HT) are discussed based on the Wavelet Transform theory. A system with two channels at resolution 2-1 and 2-2 for detecting paroxysm-spike in human's EEG is presented according to the multiresolution properties of the HT. The system adopts a coarse-to-fine strategy. First, it performs the coarse recognition on the 2-2 channel for selecting the candidate of spike in terms of rather relaxed criterion. Then, if the candidate appears, the fine recognition on the 2-1 channel is carried out for detecting spike in terms of stricter criterion. Three features of spike are extracted by investigating its intrinsic properties based on the HT. In the case of having no knowledge of prior probability of the presence of spike, the Neyman-Pearson criteria is applied to determining thresholds on the basis of the probability distribution of background and spike obtained by the results of statistical analysis to minimize error probability. The HT coefficients at resolution 2-2 and 2-1 can be computed individually and the data are compressed with 4:1 and 2:1 respectively. A half wave is regarded as the basic recognition unit so as to be capable of detecting negative and positive spikes simultaneously. The system provides a means of pattern recognition for non-stationary signal including sharp variation points in the transform domain. It is specially suitable and efficient to recognize the transient wave with small probability of occurrence in non-stationary signal. The practical examples show the performance of the system.

This paper discusses the relationships of two important program classes of linearly recursive programs, that is, decomposable programs and rule commutative programs. We prove that the decomposable programs are always rule commutative. Furthermore, the rule commutative programs that satisfy certain conditions are decomposable. These results are meaningful for integrating the related specified optimization algorithms.

An implementation of photometric stereo is described in which all directions of illumination are close to and rotationally symmetric about the viewing direction. THis has practical value but gives rise to a problem that is numerically ill-conditioned. Ill-conditioning is overcome in two ways. First, many more than the theoretical minimum number of images are acquired. Second, principal components analysis (PCA) is used as a linear preprocessing technique to determine a reduced dimensionality subspace to use as input. The approach is empirical. The ability of a radial basis function (RBF) neural network to do non-parametric functional approximation is exploited. One network maps image irradiance to surface normal. A second network maps surface normal to image irradiance. The two networks are trained using samples from a calibration sphere. Comparison between the actual input and the inversely predicted input is used as a confidence estimate. Results on real data are demonstrated.

This paper proposes a new method to determine the shape of a surface by learning the mapping between three image irradiances observed under illumination from three lighting directions and the corresponding surface gradient. The method uses Phong reflectance function to describe specular reflectance. Lambertian reflectance is included as a special case. A neural network is constructed to estimate the values of reflectance parameters and the object surface gradient distribution under the assumption that the values of reflectance parameters are not known in advance. The method reconstructs the surface gradient distribution after determining the values of reflectance parameters of a test object using two step neural network which consists of one to extract two gradient parameters from three image irradiances and its inverse one. The effectiveness of this proposed neural network is confirmed by computer simulations and by experiment with a real object.