※一部利用できない機能があります
| 1.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
This paper presents efficient implementa- tion of RLS-based adaptive filters with a large number of taps on nVIDIA GeForce graphics processing unit (GPU) and CUDA software development environment. Modification of the order and the
…
combination of calcu- lations reduces the number of accesses to slow off-chip memory. Assigning tasks into multiple threads also takes memory access order into account. For a 4096-tap case, a GPU program is almost three times faster than a CPU program.
続きを見る
|
||||
| 2.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
This paper presents efficient implementa- tion of RLS-based adaptive filters with a large number of taps on nVIDIA GeForce graphics processing unit (GPU) and CUDA software development environment. Modification of the order and the
…
combination of calcu- lations reduces the number of accesses to slow off-chip memory. Assigning tasks into multiple threads also takes memory access order into account. Multiple shader pro- cessor arrays are used to handle a large matrix. For a 8192-tap case, a GPU program is almost 30-times faster than a CPU program. Real-time processing is possible for an 8kHz-sampling and 512-tap case by us- ing 32 shader processors, which is only 25% of GeForce 8800GTS.
続きを見る
|
||||
| 3.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A simultaneous frequency- and time-domain approximation method for discrete-time
…
filters is proposed. In the method, transfer function coefficients are divided into two subsets, X//1 and X//2, which are employed for optimizing a time response and a frequency response, respectively. Frequency and time responses are optimized through the iterative Chebyshev approximation method and a method of solving linear equations, respectively. At the rth iteration step, the maximum frequency response error, which appeared at the (r minus 1)th step, is minimized, and X//2**(**r** minus **1**) becomes X//2**(**r**). X//1**(**r**) is obtained from linear equations including X//2**(**r**) as a constant. The frequency response at the rth step is evaluated using the above obtained X//1**(**r**) and X//2**(**r**). This means the optimum time response is always guaranteed in the frequency-response approximation procedure. A design example of a symmetrical impulse response shows the new approach is more efficient than conventional methods from the filter order reduction viewpoint.
続きを見る
|
||||
| 4.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科知能情報・数理<br />A training data selection method is proposed for multilayer neural networks (MLNNs). This met
…
hod selects a small number of the training data, which guarantee both generalization and fast training of the MLNNs applied to pattern classification. The generalization will be satisfied using the data locate close to the boundary of the pattern classes. However, if these data are only used in the training, convergence is slow. This phenomenon is analyzed in this paper. Therefore, in the proposed method, the MLNN is first trained using some number of the data, which are randomly selected (Step 1). The data, for which the output error is relatively large, are selected. Furthermore, they are paired with the nearest data belong to the different class. The newly selected data are further paired with the nearest data. Finally, pairs of the data, which locate close to the boundary, can be found. Using these pairs of the data, the MLNNs are further trained (Step 2). Since, there are some variations to combine Steps 1 and 2, the proposed method can be applied to both off-line and on-line training. The proposed method can reduce the number of the training data, at the same time, can hasten the training. Usefulness is confirmed through computer simulation.
続きを見る
|
||||
| 5.
論文 |
論文
|
Nakayama, Kenji ; Kato, Takuo ; Katayama, Hiroshi
|
||||
| 6.
論文 |
論文
|
Nakayama, Kenji
概要:
A bandlimited signal extrapolation algorithm is proposed. J. A. Cadzow's algorithm (1979) is modified to eliminate undesired outband spectra. An inverse filter stopband response is relaxed by adding small random numbers. A
…
constrained heuristic optimization is suggested that can use arbitrary signal properties as constraints. Through numerical examples, it is shown how regularization techniques can improve SNR (signal-to-noise-ratio) by 20 dB.
続きを見る
|
||||
| 7.
論文 |
論文
|
Keeni, Kanad ; Shimodaira, Hiroshi ; Nakayama, Kenji
概要:
This paper presents an automatic coding scheme for representing the output layer of a neural network. Compared to local representation where the number of output unit is p, the number of output unit required for the proposed representation is
…
close to log p. The output of seven different printers were used for evaluating the performance of the system. The proposed automatic representation gave the average recognition rate of 98.7% for 71 categories.
続きを見る
|
||||
| 8.
論文 |
論文
|
Wang, Youhua ; Ikeda, Kazushi ; Nakayama, Kenji
|
||||
| 9.
論文 |
論文
|
Nakayama, Kenji ; Katayama, Hiroshi
|
||||
| 10.
論文 |
論文
|
Ikeda, Kazushi ; Suzuki, Akihiro ; Nakayama, Kenji
概要:
The effects of the quantization of the parameters of a learning machine are discussed. The learning coefficient should be as small as possible for a better estimate of parameters. On the other hand, when the parameters are quantized, it
…
should be relatively larger in order to avoid the paralysis of learning originated from the quantization. How to choose the learning coefficient is given in this paper from the statistical point of view.
続きを見る
|
||||
| 11.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学大学院自然科学研究科情報システム<br />Feed-Forward (FF-) and Feed-Back (FB-) structures have been proposed for Blind Source Separati
…
on (BSS). The FF-BSS systems have some degrees of freedom in the solution space, and signal distortion is likely to occur in convolutive mixtures. On the other hand, the FB-BSS structure does not cause signal distortion. However, it requires a condition on the propagation delays in the mixing process. In this paper, source separation performance in the FB-BSS is theoretically analyzed taking the propagation delays into account. Simulation is carried out by using white signals and speech signals as the signal sources. The FF-BSS system and the FB-BSS system are compared. Even though the FB-BSS can provide good separation performance, there exits some limitation on location of the signal sources and the sensors. © Springer-Verlag Berlin Heidelberg 2007.
続きを見る
|
||||
| 12.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Arai, Shinya ; Deguchi, Masayuki
概要:
金沢大学大学院自然科学研究科知能情報・数理
|
||||
| 13.
論文 |
論文
|
Nakayama, Kenji ; Imai, Kunihiko
概要:
金沢大学理工研究域電子情報学系<br />A neural demodulator is proposed for amplitude shift keying (ASK) signal. It has several important
…
features compared with conventional linear methods. First, necessary functions for ASK demodulation, including wide-band noise rejection, pule waveform shaping, and decoding, can be embodied in a single neural network. This means these functions are not separately designed but unified in a learning and organizing process. Second, these functions can be self-organized through the learning. Supervised learning algorithms, such as the backpropagation algorithm, can be applied for this purpose. Finally, both wide-band noise rejection and a very sharp waveform response can be simultaneously achieved. It is very difficult to be done by linear filtering. Computer simulation demonstrates efficiency of the proposed method.
続きを見る
|
||||
| 14.
論文 |
論文
|
Miyoshi, Seiji ; Nakayama, Kenji
概要:
In this paper, the geometric learning algorithm (GLA) is proposed for an elementary perceptron which includes a single output neuron. The GLA is a modified version of the affine projection algorithm (APA) for adaptive filters.
…
The weights update vector is determined geometrically towards the intersection of the k hyperplanes which are perpendicular to patterns to be classified. k is the order of the GLA. In the case of the APA, the target of the coefficients update is a single point which corresponds to the best identification of the unknown system. On the other hand, in the case of the GLA, the target of the weights update is an area, in which all the given patterns are classified correctly. Thus, their convergence conditions are different. In this paper, the convergence condition of the 1st order GLA for 2 patterns is theoretically derived. The new concept `the angle of the solution area' is introduced. The computer simulation results support that this new concept is a good estimation of the convergence properties.
続きを見る
|
||||
| 15.
論文 |
論文
|
Ma, Zhiqiang ; Nakayama, Kenji ; Yamamoto, G.
|
||||
| 16.
論文 |
論文
|
Nakayama, Kenji ; Kuriki, Satoshi
概要:
A simplified digital tuned-circuit extractor is proposed. An A/D converter and multipliers are not required. An A/D converter and multipliers are not required. The tuning frequency is determined only by a master clock. The tank circuit
…
output noise is sufficiently suppressed by newly introduced error canceller. Although sampling frequency for the tank circuit is relatively low, the phase of the output signal is effectively detected. A computer simulation shows that phase adjusting for data sampling clock is stable. The proposed timing extractor is easily realized on digital integrated circuits.
続きを見る
|
||||
| 17.
論文 |
論文
|
Nakayama, Kenji
概要:
A new difference coefficient FIR digital filter is proposed. Its coefficient is obtained as the difference between the successive values of the original coefficients permuted in large magnitude sequence. It is effectively applied to
…
every kind of filter responses. Computation complexity becomes about 18% and 13% for 99th and 299th order FIR filters, compared with direct FIR filter realization.
続きを見る
|
||||
| 18.
論文 |
論文
|
Miyoshi, Seiji ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />In this paper, probabilistic memory capacity of recurrent neural networks(RNNs) is investigated. Th
…
is probabilistic capacity is determined uniquely if the network architecture and the number of patterns to be memorized are fixed. It is independent from a learning method and the network dynamics. It provides the upper bound of the memory capacity by any learning algorithms in memorizing random patterns. It is assumed that the network consists of N units, which take two states. Thus, the total number of patterns is the Nth power of 2. The probabilities are obtained by discriminations whether the connection weights, which can store random M patterns at equilibrium states, exist or not. A theoretical way for this purpose is derived, and actual calculation is executed by the Monte Carlo method. The probabilistic memory capacity is very important in applying the RNNs to real fields, and in evaluating goodness of learning algorithms. As an example of a learning algorithm, the improved error correction learning is investigated, and its convergence probabilities are compared with the upper bound. A linear programming method can be effectively applied to this numerical analysis.
続きを見る
|
||||
| 19.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
Nowadays, the integer prime-factorization problem finds its application often in modern cryptography. Artificial Neural
…
Networks (ANNs) have been applied to the integer prime-factorization problem. A composed number N is applied to the ANNs, and one of its prime factors p is obtained as the output. Previously, neural networks dealing with the input and output data in a decimal format have been proposed. However, accuracy is not sufficient. In this paper, a neural network following a binary approach is proposed. The input N as well as the desired output p were expressed in a binary form. The proposed neural network is expected to be more stable, i.e. less sensitive to small errors in the network outputs. Simulations have been performed and the results are compared with the results reported in the previous study. The number of required search times for the true prime number can be well reduced. Furthermore, the probability density function of the training patterns is investigated and the need for different data creation and/or selection techniques is shown. © 2005 IEEE.
続きを見る
|
||||
| 20.
論文 |
論文
|
Nakayama, Kenji ; Ikehara, Keisuke
|
||||
| 21.
論文 |
論文
|
Nakayama, Kenji ; Kimura, Yoshinori ; Katayama, Hiroshi
概要:
In this paper, quantization level increase in human face images using a multilayer neural network (NN) is investigated.
…
Basically speaking, it is impossible to increase quality without any other information. However, when images are limited to some category, image restoration could be possible, based on the common properties in this category. The multilayer NN is trained using human face images of 32ﾃ・2 pixels with 8-levels as the input data, and 256-level images as the targets. The standard back-propagation (BP) algorithm is employed. 20, 40 and 100 training data are examined. By increasing the training data, a general function of regenerating missing information can be achieved. The internal structure of the trained NN is analyzed using some special input images. As a result, it has been confirmed that the NN regards the input image as the human face, and extracts features of the face. The input image is transformed using these features and the common properties of the training data, extracted and held on the connection weights, to the human face image.
続きを見る
|
||||
| 22.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Katoh, Shinya ; Yamamoto, Tadashi ; Nakanishi, Kenichi ; Sawada, Manabu
概要:
In training neural networks, it is important to reduce input variables for saving memory, reducing network size, and ach
…
ieving fast training. This paper proposes two kinds of selecting methods for useful input variables. One of them is to use information of connection weights after training. If a sum of absolute value of the connection weights related to the input node is large, then this input variable is selected. In some case, only positive connection weights are taken into account. The other method is based on correlation coefficients among the input variables. If a time series of the input variable can be obtained by amplifying and shifting that of another input variable, then the former can be absorbed in the latter. These analysis methods are applied to predicting cutting error caused by thermal expansion and compression in machine tools. The input variables are reduced from 32 points to 16 points, while maintaining good prediction within 6 ﾎｼm, which can be applicable to real machine tools.
続きを見る
|
||||
| 23.
論文 |
論文
|
Kuraishi, Yoshiaki ; Makabe, Takayoshi ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />This paper presents a fully integrated analog front-end LSI chip which is an inte
…
rface system between digital signal processors and existing analog telecommunication networks. The developed analog LSI chip includes many high level function blocks such as A/D and D/A converters with 11 bit resolution, various kinds of SCF's, an agc circuit, an external control level adjuster, a carrier detector, and a zero crossing detector. Design techniques employed are mainly directed toward circuit size reductions. The LSI chip is fabricated in a 5 mu m line double polysilicon gate NMOS process. Chip size is 7. 14 multiplied by 6. 51 mm. The circuit operates on plus or minus 5 v power supplies. Typical power consumption is 270 mw. By using this analog front-end LSI chip and a digital signal processor, modern systems can be successfully constructed in a compact size.
続きを見る
|
||||
| 24.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
This paper analyzes signal separation and distortion in a feedforward blind source separation (BSS) applied to convolutive mixture environment. In the BSS learning, a separation block is adjusted so as to make the output signals
…
statistically independent. One direction for this adjustment is to extract only one signal source at a single output. The other direction is to make the output signal spectra white. The second direction causes signal distortion and conflicts with the first direction. This confliction prevent signal separation. These relations are also dependent on a length of the impulse responses in the mixture and the filter length in the separation block. These properties are theoretically analyzed and are investigated through simulations using the white signal sources and the speech signal sources.
続きを見る
|
||||
| 25.
論文 |
論文
|
Xu, Q. ; Nakayama, Kenji
概要:
This paper investigates some possible problems of Cascade Correlation algorithm, one of which is the zigzag output mapping caused by weight-illgrowth of the adding hidden unit. Without doubt, it could lead to deteriorate the
…
generalization, especially for regression problems. To solve this problem, we combine Cascade Correlation algorithm with regularization theory. In addition, some new regularization terms are proposed in light of special cascade structure. Simulation has shown that regularization indeed smooth the zigzag out-put, so that the generalization is improved, especially for functional approximation.
続きを見る
|
||||
| 26.
論文 |
論文
|
Nakayama, Kenji ; Chigawa, Yasuhide ; Hasegawa, Osamu
|
||||
| 27.
論文 |
論文
|
27. A weighted competitive learning method extracting skeleton pattern for Japanese Kanji characters
Nakayama, Kenji ; Kato, Takuo
|
||||
| 28.
論文 |
論文
|
Tokui, N. ; Nakayama, Kenji ; Hirano, Akihiro
概要:
In order to achieve fast convergence and less computation for adaptive filters, a joint method combining a whitening process and the NLMS algorithm is a hopeful approach. However, updating the filter coefficients is not synchronized with the
…
reflection coefficient updating resulting in unstable behavior. We analyzed effects of this, and proposed the Synchronized Learning Algorithm to solve this problem. Asynchronous error between them is removed, and fast convergence and small residual error were obtained. This algorithm, however, requires O(ML) computations, where M is an adaptive filter length, and L is a lattice predictor length. It is still large compared with the NLMS algorithm. In order to achieve less computation while the fast convergence is maintained, a block implementation method is proposed. The reflection coefficients are updated at some period, and are fixed during this interval. The proposed block implementation can be effectively applied to parallel form adaptive filters, such as sub-band adaptive filters. Simulation using speech signal shows that a learning curve of the proposed block implementation a little slower than the our original algorithm, but can save the computational complexity.
続きを見る
|
||||
| 29.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Ushimaru, S.
|
||||
| 30.
論文 |
論文
|
Yukawa, Akira ; Maruta, Rikio ; Nakayama, Kenji
|
||||
| 31.
論文 |
論文
|
Nakayama, Kenji ; Iwata, Atsushi ; Yanagisawa, Takeshi
概要:
金沢大学大学院自然科学研究科情報システム
|
||||
| 32.
論文 |
論文
|
Hibino, Masao ; Nakayama, Kenji ; Mizukami, Toshihiko
|
||||
| 33.
論文 |
論文
|
Ohnishi, K. ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />A neural demodulator is proposed for quadrature amplitude modulation (QAM) signals. It has several
…
important features compared with conventional linear methods. First, necessary functions for the QAM demodulation, including wide-band noise rejection, pulse waveform shaping, and decoding, can be embedded in a single neural network. This means that these functions are not separately designed but are unified in a learning process. Second, these functions can be self-organized through the learning. Supervised learning algorithms, such as the back-propagation algorithm, can be applied for this purpose. Finally, both wide-band noise rejection and a very sharp waveform response can be simultaneously achieved. It is very difficult to be done by linear filtering. Computer simulation demonstrates efficiency of the proposed method.
続きを見る
|
||||
| 34.
論文 |
論文
|
Wang, Youhua ; Terada, Yasuhiro ; Matsui, Minoru ; Iida, Kazuhiro ; Nakayama, Kenji
|
||||
| 35.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />A training data reduction method for a multilayer neural network (MLNN) is proposed in this paper.
…
This method reduce the data by selecting the minimum number of training data that guarantee generality of the MLNN. For this purpose, two methods are used. One of them is a pairing method which selects the training data by finding the nearest data of the different classes. Data along the class boundary in data space can be selected. The other method is a training method, which used a semi-optimum MLNN in a training process. Since the MLNN classify data based on the distance from the network boundary, the selected data can locate close to the class boundary. So, if the semi-optimum MLNN did not select data from class boundary, pairing method can select them. The proposed methods can be applied to both off-line training and on-line training. The proposed method is also investigated through computer simulation.
続きを見る
|
||||
| 36.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
概要:
Signal classification performance using multilayer neural network (MLNN) and the conventional signal processing methods are theoretically compared under the limited observation period and computational load. The signals with N samples
…
are classified based on frequency components. The comparison is carried out based on degree of freedom the signal detection regions in an N-dimensional signal space. As a result, the MLNN has higher degree of freedom, and can provide more flexible performance for classifying the signals than the conventional methods. This analysis is further investigated throught computer simulations. Multi-frequency signals and the real application, a dial tone receiver, are taken into account. As a result, the MLNN can provide much higher accuracy than the conventional signal processing methods.
続きを見る
|
||||
| 37.
論文 |
論文
|
Nakayama, Kenji ; Katou, Haruo ; Hirano, Akihiro
|
||||
| 38.
論文 |
論文
|
Nakayama, Kenji ; Sato, Yayoi ; Kuraishi, Yoshiaki
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />Design techniques are described for a switched capacitor adaptive line equalizer
…
which is applied to high-speed (200 kb/s) digital transmission over analog subscriber loops. An equalizer transfer function is approximated so as to minimize intersymbol interference of an isolated pulse response. Optimum pole-zero location, which is suited to line characteristics in a wide frequency band, is also discussed. In order to attain high accuracy capacitor ratios using a small unit capacitor, capacitor values are rounded off into equivalent integer values, and are discretely optimized using pole-zero deviation as an error criterion. The equalizer has a finite number of frequency responses which correspond to line lengths. Gain and delay time differences between the adjoining step responses are compressed. The switched capacitor line equalizer was fabricated using 3- mu m CMOS technology. Measured data were very close to designed performances.
続きを見る
|
||||
| 39.
論文 |
論文
|
Nakayama, Kenji ; Mizukami, Toshihiko
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A new infinite impulse response (IIR) Nyquist filter with zero intersymbol interf
…
erence is proposed. The necessary and sufficient conditions for the transfer function are obtained. The proposed IIR Nyquist filter requires only frequency-domain optimization. Multistep optimization, using the iterative Chebyshev approximation, is proposed. This method is able to design a new kind of IIR Nyquist filter with the minimum order. Numerical examples for 30- and 15-percent rolloff rates are illustrated. From these examples, it is confirmed that the IIR approach can reduce the filter order and hardware size, compared with the conventional finite impulse response (FIR) Nyquist filters. Its efficiency becomes marked for high Q Nyquist filters.
続きを見る
|
||||
| 40.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />This paper proposes a new discrete optimization method which is mainly directed toward saving
…
computing time for high-order FIR filters. A transfer functions first approximated in a cascade form of a low-order function W(z) with pre-rounded coefficients and a high-order function F(z) with infinite precision coefficients. Rounded F(z) coefficients are discretely optimized to minimize the mean square error of the amplitude response. The roundoff error spectrum is shaped to be suppressed by a weighting function W(z). In order to save computing time, the error is equivalently evaluated in a time domain, and the F(z) coefficients are divided into small groups in the discrete optimization procedure. Design examples for 200 tap FIR filters demonstrate practical usefulness.
続きを見る
|
||||
| 41.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
Source separation and signal distortion in three kinds of BSSs with convolutive mixture are analyzed. They include a fee
…
dforward BSS, trained in the time domain and in the frequency domain, and a feedback BSS, trained in the time domain. First, an evaluation measure of signal distortion is discussed. Second, conditions for source separation and distortion free are derived. Based on these conditions, source separation and signal distortion are analyzed. The feedforward BSS has some degree of freedom, and the output spectrum can be changed. The feedforward BSS, trained in the frequency domain, has weighting effect, which can suppress signal distortion. This weighting is, however, effective only when the source spectra are similar to each other. Since, the feedforward BSS, trained in the time domain, does not have any constraints on signal distortion free, its output signals can be easily distorted. A new learning algorithm with a distortion free constraint is proposed. On the other hand, the feedback BSS can satisfy both source separation and distortion free conditions simultaneously. Simulation results support the theoretical analysis. © 2005 IEEE.
続きを見る
|
||||
| 42.
論文 |
論文
|
Miyoshi, Seiji ; Ikeda, Kazushi ; Nakayama, Kenji
|
||||
| 43.
論文 |
論文
|
Nakayama, Kenji
|
||||
| 44.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Watanabe, Kazunobu
概要:
This paper presents convergence characteristics of stereophonic echo cancellers with pre-processing. The convergence analysis of the averaged tap-weights show that the convergence characteristics depends on the relation between the impulse
…
response in the far-end room and the changes of the pre-processing filters. Examining the uniqueness of the solution in the frequency domain leads us to the same relation. Computer simulation results show the validity of these analyses.
続きを見る
|
||||
| 45.
論文 |
論文
|
Nakayama, Kenji ; Ohsugi, M.
|
||||
| 46.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakaguchi, Hiroaki
|
||||
| 47.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Nishiwaki, Takayuki
概要:
A network structure and its learning algorithm have been proposed for blind source separation applied to nonlinear mixtures. The network has a cascade form consists of a source separation block and a linearization block in this order.
…
The conventional learning algorithm is employed for the separation block. A new learning algorithm is proposed for the linearization block assuming 2nd-order nonlinearity. After, source separation, the outputs include the nonlinear components for the same signal source. This nonlinearity is suppressed through the linearization block. Parameters in this block are iteratively adjusted based on a process of solving a 2nd-order equation of a single variable. Simulation results, using 2-channel speech signals and an instantaneous nonlinear mixing process, show good separation performance.
続きを見る
|
||||
| 48.
論文 |
論文
|
Nakayama, Kenji ; Mitsutani, N.
概要:
Summary form only given. A novel adaptive threshold method is proposed for an associative memory using a mutually connec
…
ted neural network. Computer simulation was carried out, using 51 and 153 patterns, which appear on a keyboard. The neural network has 16 × 16 = 256 units and full connections. Results demonstrate that dramatic improvements in memory capacity and association rates can be achieved. For example, an association rate for 51 patterns with 40 noises has been increased from 12.2% to 97.7%, compared with a single threshold method.
続きを見る
|
||||
| 49.
論文 |
論文
|
Nakayama, Kenji ; Mizukami, T.
概要:
A new approach for low-bit-rate coding is proposed. Numerical evaluation for the bit-rate compression ratio and SNR shows the proposed method provides good performance compared with conventional linear PCM coding. Furthermore, since the
…
predictor in the coder does not utilize any of the input signal statistics, this method can be applied for both speech and music coding.
続きを見る
|
||||
| 50.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
概要:
Frequency analysis capability of multilayer neural networks, trained by back-propagation (BP) algorithm is investigated. Multi-frequency signal classification is taken into account for this purpose. The number of frequency
…
sets, that is signal groups, is 2approx.5, and the number of frequencies included in a signal group is 3approx.5. The frequencies are alternately located among the signal groups. Through computer simulation, it has been confirmed that the neural network has very high resolution. Classification rates are about 99.5% for training signals, and 99.0% for untraining signals. The results are compared with conventional methods, including Euclidean distance with accuracy of about 65%, Fourier transform with accuracy of about 10approx.30%, and using very high-Q filters with a huge number of computations. The neural network requires only the same number of inner products as the hidden units. Frequency sensitivity and robustness for the random noise are studied. The networks show high frequency sensitivity, namely, the networks have high frequency resolution. Random noise are added to the multi-frequency signals to investigate how does the network cancel uncorrelated noise among the signals. By increasing the number of samples, or training signals, effects of random noise can be cancelled.
続きを見る
|
||||
| 51.
論文 |
論文
|
Khalaf, Ashraf A.M. ; Nakayama, Kenji
|
||||
| 52.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Fusakawa, M.
概要:
In multilayer neural networks, network size reduction and fast convergence are important. For this purpose, trainable activation functions and nonlinear synapses have been proposed. When high-order polynomials are used for
…
nonlinearity, the number of terms in the polynomial becomes a very large for a high-dimensional input. It causes very complicated networks and slow convergence. In this paper, a method to select the useful terms in the polynomial in a learning process is proposed. This method is based on the genetic algorithm (GA), and combines the internal information, magnitude of connection weihgts, to select the gene in the next generation. A mechanism of pruning the terms is inherently included. Many examples demonstrate usefulness of the proposed method compared with the ordinary GA method. Convergence is stable and the number of the selected terms is well reduced.
続きを見る
|
||||
| 53.
論文 |
論文
|
Nakayama, Kenji ; Takahashi, Yutaka ; Sato, Yayoi ; Nukada, Yasuaki
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />An adaptive switched-capacitor (SC) line equalizer system that can be applied to
…
four-wire full-duplex and multirate digital transmission is described. Several kinds of noises can exist in programmable high-gain SC equalizers, such as switching noise, DC offset jump, and transient response. To avoid their effects, an adaptive SC filter is proposed. Two identical SC circuits are used in parallel. An input signal is fed into both SC circuits, and output signals from both circuits are alternately sent after the above noises are eliminated. Furthermore, many different data rates can be handled by slightly modifying an equalizer circuit. A bridged-tap echo canceller and a DC offset canceller are modified so as to be applied to the proposed adaptive SC filter. A line equalizer system, that handles data rates ranging from 3.2 to 64 kb/s was designed. An LSI was fabricated using a 3-μm CMOS process. Experimental results show that noise effects are mostly eliminated, and frequency responses and eye openings are close to the designed values.
続きを見る
|
||||
| 54.
論文 |
論文
|
Khalaf, Ashraf A.M. ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />Time series prediction is very important technology in a wide variety of fields. The actual ti
…
me series contains both linear and nonlinear properties. The amplitude of the time series to be predicted is usually continuous value. For these reasons, we combine nonlinear and linear predictors in a cascade form. The nonlinear prediction problem is reduced to a pattern classification. A set of the past samples x(n - 1), . . . , x(n - N) is transformed into the output, which is the prediction of the next coming sample x(n). So, we employ a multi-layer neural network with a sigmoidal hidden layer and a single linear output neuron for the nonlinear prediction. It is called a Nonlinear Sub-Predictor (NSP). The NSP is trained by the supervised learning algorithm using the sample x(n) as a target. However, it is rather difficult to generate the continuous amplitude and to predict linear property. So, we employ a linear predictor after the NSP. An FIR filter is used for this purpose, which is called a Linear Sub-Predictor (LSP). The LSP is trained by the supervised learning algorithm using also i(n) as a target. In order to estimate the minimum size of the proposed predictor, we analyze the nonlinearity of the time series of interest. The prediction is equal to mapping a set of past samples to the next coming sample. The multi-layer neural network is good for this kind of pattern mapping. Still, difficult mappings may exist when several sets of very similar patterns are mapped onto very different samples. The degree of difficulty of the mapping is closely related to the nonlinearity. The necessary number of the past samples used for prediction is determined by this nonlinearity. The difficult mapping requires a large number of the past samples. Computer simulations using the sunspot data and the artificially generated discrete amplitude data have demonstrated the efficiency of the proposed predictor and the nonlinearity analysis.
続きを見る
|
||||
| 55.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Kanbe, Aki
概要:
Many problems solved by multilayer neural networks (MLNNs) are reduced into pattern mapping. If the mapping includes several different rules, it is difficult to solve these problems by using a single MLNN with linear connection
…
weights and continuous activation functions. In this paper, a structure trainable neural network has been proposed. The gate units are embedded, which can be trained together with the connection weights. Pattern mapping problems, which include several different mapping rules, can be realized using a single new network. Since, some parts of the network can be commonly used for different mapping rules, the network size can be reduced compared with the modular neural networks, which consists of several independent expert networks.
続きを見る
|
||||
| 56.
論文 |
論文
|
Tokui, N. ; Nakayama, Kenji ; Hirano, Akihiro
|
||||
| 57.
論文 |
論文
|
Miyoshi, Seiji ; Nakayama, Kenji
概要:
A recurrent neural network (RNN), in which each unit has serial delay elements, is proposed for memorizing limit cycles (LCs). This network is called DRNN in this paper. An LC consists of several basic patterns. The hysteresis information
…
of LCs, realized on the connections from the delay elements to the units, is very efficient in the following reasons. First, the same basic patterns can be shared by different LCs. This make it possible to drastically increase the number of LCs, even though using a small number of the basic patterns. Second, noise performance, that is, probability of recalling the exact LC starting from the noisy LC, can be improved. The hysteresis information consists of two components, the order of the basic patterns included in an LC, and the cross-correlation among all the basic patterns. The former is highly dependent on the number of LCs, and the latter the number of all the basic patterns. In order to achieve good noise performance, a small number of the basic patterns is preferred. These properties of the DRNN are theoretically analyzed and confirmed through computer simulations. It is also confirmed that the DRNN is superior to the RNN without delay elements for memorizing LCs.
続きを見る
|
||||
| 58.
論文 |
論文
|
Ma, Zhiqiang ; Ikeda, Tetsuya ; Nakayama, Kenji
|
||||
| 59.
論文 |
論文
|
Nakayama, Kenji ; Inomata, Satoru ; Takeuchi, Yukou
概要:
金沢大学大学院自然科学研究科情報システム<br />A design methodology is presented for digital multilayer neural networks with limited binary e
…
xpressions. An error back-propagation algorithm is modified as follows: the numbers of binary bits used for connections and unit outputs are decreased step by step in the training process. In order to express unit outputs with two-level values, the differential of the logistic function is replaced by a small positive constant used in weight change equations. After the training is completed, binary expressions for connections and unit outputs can be reduced to several-bit and two-level values, respectively. Therefore, no multipliers or nonlinear functions are required in the resulting network, which will be used for pattern recognition. Furthermore, memory capacity and adder circuit hardware can be reduced. The network performance is also insensitive to noisy patterns.
続きを見る
|
||||
| 60.
論文 |
論文
|
Kobori, Hideki ; Ikeda, Kazushi ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />A model of dynamic associative memories is proposed in this paper. The aim is to find all stored pa
…
tterns, and to distinguish the stored and the spurious patterns. Aihara used chaotic neurons and showed that his model has a nonperiodic associative dynamics. In his model, however, it is difficult to distinguish the stored patterns from the others, because the state of the network changes continually. We propose such a new model of neurons that each neuron changes its output to the other when the accumulation of its internal state exceeds a certain threshold. By computer experiments, we show that the state of the network stays at the stored pattern for a while and then travels around to another pattern, and so on. Furthermore, when the number of the stored patterns is small, the stored and the spurious patterns can be distinguished using interval of the network staying these patterns.
続きを見る
|
||||
| 61.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />An improved FFT (fast Fourier transform) algorithm combining both decimations
…
in frequency and in time is presented. Stress is placed on a derivation of general formulas for submatrices and multiplicands. Computational efficiency is briefly discussed.
続きを見る
|
||||
| 62.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A new approach to the implmentation of sum of products (SP) and transposed sum of
…
products (TSP) using partitioned arithmetic is proposed. This aproach makes it possible to realize SP and TSP with recent advanced semiconductor memory technology without multipliers. It is also shown to offer signficant reductions in memory capacity requirement and computational complexity. Dynamic range constraints and output roundoff noise analysis are also discussed.
続きを見る
|
||||
| 63.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
金沢大学大学院理工研究域電子情報学系<br />OveOver the years, many improvements and refinements of the backpropagation learning algorithm h
…
ave been reported. In this paper, a new adaptive penalty-based learning extension for the backpropagation learning algorithm and its variants is proposed. The new method initially puts pressure on artificial neural networks in order to get all outputs for all training patterns into the correct half of the output range, instead of mainly focusing on minimizing the difference between the target and actual output values. The technique is easy to implement and computationally inexpensive. In this study, the new approach has been applied to the backpropagation learning algorithm as well as the RPROP learning algorithm and simulations have been performed. The superiority of the new proposed method is demonstrated. By applying the extension, the number of successful runs can be greatly increased and the average number of epochs to convergence can be well reduced on various problem instances. Furthermore, the change of the penalty values during training has been studied and its observation shows the active role the penalties play within the learning process. © 2006 IEEE.
続きを見る
|
||||
| 64.
論文 |
論文
|
Nakayama, Kenji ; Nishimura, Katsuaki ; Katayama, Hiroshi
概要:
Authors have proposed an asymmetrical associative neural network (NN) using variable hysteresis threshold and its learning and association algorithms. It can drastically improve noise performance, that is insensitivity to noise. In
…
this paper, memory capacity bound and threshold optimization in this associative NN are further discussed. Binary random patterns are considered. First, relation between the number of patterns and the number of iterations is investigated. The latter gradually increases until some number of patterns. After that, it suddenly increases. This is a very peculiar phenomenon. This turning point gives the memory capacity bound, that is about 1.56N, where N is the number of units. Next, threshold optimization is discussed. Relation between threshold and noise performance, and effects of connection weight distribution on noise performance are theoretically discussed. Based on these results, a ratio of step-size and the threshold is optimized to be 0.5/(NP-1), where NP is the number of units on the pattern. Numerically statistical simulation demonstrates efficiency of the proposed methods.
続きを見る
|
||||
| 65.
論文 |
論文
|
Khalaf, Ashraf A.M. ; Nakayama, Kenji
概要:
Time series prediction is a very important technology in a wide variety of field. The actual time series contains both linear and nonlinear properties. The amplitude of the time series to be predicted is usually continuous value. For
…
this reason, we combine nonlinear and linear predictors in a cascade form. In order to estimate the minimum size of the proposed predictor, we propose a nonlinearity analysis for the time series of interest. Computer simulations using the sunspot data have demonstrated the efficiency of the proposed predictor and the nonlinearity analysis.
続きを見る
|
||||
| 66.
論文 |
論文
|
Nakayama, Kenji ; Ma, Zhiqiang ; Yamamoto, G.
概要:
A new pole-zero pairing and ordering strategy which can provide the minimum total capacitance in cascaded biquad switched-capacitor (SC) filters is proposed. Design examples for several kinds of SC filters are demonstrated. The
…
proposed conditions can limit pairing and ordering assignments, by which the total capacitance is concentrated within 5% over the minimum value.
続きを見る
|
||||
| 67.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学大学院自然科学研究科<br />Source separation and signal distortion are theoretically analyzed for the FF-BSS systems implemente
…
d in both the time and frequency domains and the FB-BSS system. The FF-BSS systems have some degree of freedom, and cause some signal distortion. The FB-BSS has a unique solution for complete separation and distortion free. Next, the condition for complete separation and distortion free is derived for the FF-BSS systems. This condition is applied to the learning algorithms. Computer simulations by using speech signals and stationary colored signals are carried out for the conventional methods and the new learning algorithms employing the proposed distortion free constraint. The proposed method can drastically suppress signal distortion, while maintaining high separation performance. The FB-BSS system also demonstrates good performances. The FF-BSS systems and the FB-BSS system are compared based on the transmission time difference in the mixing process. Location of the signal sources and the sensors are rather limited in the FB-BSS system. © 2006 IEEE.
続きを見る
|
||||
| 68.
論文 |
論文
|
Ma, Zhiqiang ; Nakayama, Kenji ; Sugiyama, Akihiko
概要:
金沢大学大学院自然科学研究科情報システム<br />An automatic tap assignment method in subband adaptive filter is proposed in this letter. The
…
number of taps of the adaptive filter in each band is controlled by the mean-squared error. The numbers of taps increase in the bands which have large errors, while they decrease in the bands having small errors, until residual errors in all the bands become the same. In this way, the number of taps in a band is roughly proportional to the length of the impulse response of the unknown system in this band. The convergence rate and the residual error are improved, in comparison with existing uniform tap assignment. Effectiveness of the proposed method has been confirmed through computer simulation.
続きを見る
|
||||
| 69.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />Over the years, many improvements and refinements to the backpropagation learning algorithm ha
…
ve been reported. In this paper, a new adaptive penalty-based learning extension for the backpropagation learning algorithm and its variants is proposed. The new method initially puts pressure on artificial neural networks in order to get all outputs for all training patterns into the correct half of the output range, instead of mainly focusing on minimizing the difference between the target and actual output values. The upper bound of the penalty values is also controlled. The technique is easy to implement and computationally inexpensive. In this study, the new approach is applied to the backpropagation learning algorithm as well as the RPROP learning algorithm. The superiority of the new proposed method is demonstrated though many simulations. By applying the extension, the percentage of successful runs can be greatly increased and the average number of epochs to convergence can be well reduced on various problem instances. The behavior of the penalty values during training is also analyzed and their active role within the learning process is confirmed. Copyright © 2006 The Institute of Electronics, Information and Communication Engineers.
続きを見る
|
||||
| 70.
論文 |
論文
|
Nakayama, Kenji ; Seki, T.
|
||||
| 71.
論文 |
論文
|
Nakayama, Kenji ; Takahashi, Yutaka ; Satoh, Yayoi ; Naka, Masahiro ; Nukada, Yasuaki
概要:
The authors present a single-chip switched capacitor (SC) line equalizer system which can be applied to full duplex digital transmission. Adaptive SC filters, constructed with programmable capacitor arrays (PCAs), usually cause undesired
…
responses, such as spike noise and transient response, which degrade data transmission quality. To avoid these phenomena, a duplex SC equalizer which has the same circuits in parallel is introduced. The PCAs in one duplex equalizer, whose output is not transferred, are varied. After the undesired phenomena vanish, the output is alternated. The equalizer system can be applied at several different bit rates by changing the external control signals. Algorithms and circuit realizations for a bridged tap echo canceller and a dc offset canceller are further improved.
続きを見る
|
||||
| 72.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Horita, Akihide
|
||||
| 73.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
|
||||
| 74.
論文 |
論文
|
Keeni, Kanad ; Nakayama, Kenji ; Shimodaira, Hiroshi
概要:
A method has been proposed for weight initialization in back-propagation feed-forward networks. Training data is analyzed and the notion of critical point is introduced for determining the initial weights and the number of hidden
…
units. The proposed method has been applied to artificial data and the publicly available cancer database. The experimental results of artificial data show that the proposed method takes 1/3 of the training time required for standard back-propagation. In order to verify the effectiveness of the proposed method, standard back-propagation, where the learning starts with random initial weights was also applied to the cancer database. The experimental results indicate that the proposed weight initialization method results in better generalization.
続きを見る
|
||||
| 75.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部
|
||||
| 76.
論文 |
論文
|
Khalaf, Ashraf A.M. ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />A nonlinear time series predictor was proposed, in which a nonlinear sub-predictor (NSP) and a
…
linear subpredictor (LSP) are combined in a cascade form. This model is called hybrid predictor here. The nonlinearity analysis method of the input time series was also proposed to estimate the network size. We have considered the nonlinear prediction problem as a pattern mapping one. A multi-layer neural network, which consists of sigmoidal hidden neurons and a single linear output neuron, has been employed as a nonlinear sub-predictor. Since the NSP includes nonlinear functions, it can predict the nonlinearity of the input time series. However, the prediction is not complete in some cases. Therefore, the NSP prediction error is further compensated for by employing a linear sub-predictor after the NSP. In this paper, the prediction mechanism and a role of the NSP and the LSP are theoretically and experimentally analyzed. The role of the NSP is to predict the nonlinear and some part of the linear property of the time series. The LSP works to predict the NSP prediction error. Furthermore, predictability of the hybrid predictor for noisy time series is investigated. The sigmoidal functions used in the NSP can suppress the noise effects by using their saturation regions. Computer simulations, using several kinds of nonlinear time series and other conventional predictor models, are demonstrated. The theoretical analysis of the predictor mechanism is confirmed through these simulations. Furthermore, predictability is improved by slightly expanding or shifting the input potential of the hidden neurons toward the saturation regions in the learning process.
続きを見る
|
||||
| 77.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A new approach is proposed for computing the discrete Fourier transform (DFT) wit
…
h power-of-2 length using the butterfly-structure number-theoretic transform (NTT). An algorithm breaking down the DFT matrix into circular matrices with power-of-2 size is introduced. Fast circular convolution, which is implemented by the NTT based on the butterfly structure, provides significant reductions in the number of computations, as well as a simple and regular structure. The proposed algorithm can be successively implemented following a simple flowchart using the reduced-size submatrices. Multiplicative complexity is reduced to about 21% of that with the classical FFT algorithm, preserving almost the same number of additions.
続きを見る
|
||||
| 78.
論文 |
論文
|
Wang, Youhua ; Ikeda, Kazushi ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />The numerical property of an adaptive filter algorithm is the most important prob
…
lem in practical applications. Most fast adaptive filter algorithms have the numerical instability problem and the fast Newton transversal filter (FNTF) algorithms are no exception. In this paper, we propose a numerically stable fast Newton type adaptive filter algorithm. Two problems are dealt with in the paper. First, we derive the proposed algorithm from the order-update fast least squares (FLS) algorithm. This derivation is direct and simple to understand. Second, we give the stability analysis using a linear time-variant state-space method. The transition matrix of the proposed algorithm is given. The eigenvalues of the ensemble average of the transition matrix are shown to be asymptotically all less than unity. This results in a much improved numerical performance compared with the FNTF algorithms. The computer simulations implemented by using a finite-precision arithmetic have confirmed the validity of our analysis.
続きを見る
|
||||
| 79.
論文 |
論文
|
Hara, Kazuyuki ; Amakata, Yoshihisa ; Nukaga, Ryohei ; Nakayama, Kenji
概要:
In this paper, we propose a learning method that updates a synaptic weight in probability which is proportional to an output error. Proposed method can reduce computational complexity of learning and at the same time, it can improve the
…
classification ability. We point out that an example produces small output error does not contribute to update of a synaptic weight. As learning progresses, the number of the small error examples will be increasing compared to the big one is decreasing. This unbalance will cause of difficulty of learning large error examples. Proposed method cancels this phenomenon and improve the learning ability. Validity of proposed method is confirmed through computer simulation.
続きを見る
|
||||
| 80.
論文 |
論文
|
Kuraishi, Yoshiaki ; Nakayama, Kenji ; Miyadera, Kazuyuki ; Okamura, Toshiyuki
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A single-chip speech spectrum analyzer which contains a 20-channel filter bank, a
…
9 bit resolution analog-to-digital converter, and a 396 bit buffer memory is described. Several efficient design techniques were used to realize the equivalent 308th-order transfer functions on a single chip monolithic MOS circuit. A new time division multiplexed switched-capacitor filter technique is introduced which can easily cancel dc offsets which appear in the multiplexed channel outputs. The LSI was fabricated in 3. 5 mu m CMOS technology, with a 7. 0 multiplied by 6. 5 mm**2 die size and a 150 mw power consumption with a plus 5 v single power supply. Experimental results show that designed performance was realized.
続きを見る
|
||||
| 81.
論文 |
論文
|
Hara, Kazuyuki ; Nakayama, Kenji
概要:
A training data selection method for multi-class data is proposed. This method can be used for multilayer neural networks (MLNN). The MLNN can be applied to pattern classification, signal process, and other problems that can be
…
considered as the classification problem. The proposed data selection algorithm selects the important data to achieve a good classification performance. However, the training using the selected data converges slowly, so we also propose an acceleration method. The proposed training method adds the randomly selected data to the boundary data. The validity of the proposed methods is confirmed through the computer simulation.
続きを見る
|
||||
| 82.
論文 |
論文
|
Nakayama, Kenji ; Kimura, Yoshinori
概要:
金沢大学理工研究域電子情報学系<br />An optimization method of activation functions is proposed. Three typical functions are combined in
…
hidden layers. Contribution of the functions is evaluated using three criteria. The useful functions are selected or multiplied in the learning process. Problems of parity and of counting '1' in bit-patterns can be solved by the proposed method with the suitable functions and the minimum number of hidden units.
続きを見る
|
||||
| 83.
論文 |
論文
|
Tomikawa, Y. ; Nakayama, Kenji
概要:
In this paper, a recurrent neural network (RNN) is applied to approximating one to N many valued mappings. The RNN described in this paper has a feedback loop from an output to an input in addition to the conventional multi layer neural
…
network (MLNN). The feedback loop causes dynamic output properties. The convergence property in these properties can be used for this approximating problem. In order to avoid conflict by the overlapped target data y*s to the same input x., the input data set (x*, y*) and the target data y* are presented to the network in learning phase. By this learning, the network function f(x, z) which satisfies y* = f(x*,y*) is formed. In recalling phase, the solutions y of y = f(x,y) are detected by the feedback dynamics of RNN. The different solutions for the same input x can be gained by changing the initial output value of y. It have been presented in our previous paper that the RNN can approximate many valued continuous mappings by introducing the differential condition to learning. However, if the mapping has discontinuity or changes of value number, it sometimes shows undesirable behavior. In this paper, the integral condition is proposed in order to prevent spurious convergence and to spread the attractive regions to the approximating points.
続きを見る
|
||||
| 84.
論文 |
論文
|
Wang, Hong ; Ma, Zhiqiang ; Nakayama, Kenji
概要:
In this paper we investigate the evolutionary heuristics used as approximation algorithm to the subset sum problem. We p
…
ropose a graded penalty function in a fitness function of genetic algorithms to penalize an infeasible string in solving the subset sum problem. An exponential term of generation variable, tθ, is added into the penalty function for increasing penalty generation by generation. The experiments show that the proposed penalty function is more efficient than other existing penalty functions. It is suggested that the penalty pressure is increased step by step.
続きを見る
|
||||
| 85.
論文 |
論文
|
Wang, Youhua ; Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />The numerical properties of the recursive least squares (RLS) algorithm and its fast versions
…
have been extensively studied. However, very few investigations are reported concerning the numerical behavior of the predictor based least squares (PLS) algorithms that provide the same least squares solutions as the RLS algorithm. This paper presents a comparative study on the numerical performances of the RLS and the backward PLS (BPLS) algorithms. Theoretical analysis of three main instability sources reported in the literature, including the over-range of the conversion factor, the loss of symmetry and the loss of positive definiteness of the inverse correlation matrix, has been done under a finite-precision arithmetic. Simulation results have confirmed the validity of our analysis. The results show that three main instability sources encountered in the RLS algorithm do not exist in the BPLS algorithm. Consequently, the BPLS algorithm provides a much more stable and robust numerical performance compared with the RLS algorithm.
続きを見る
|
||||
| 86.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Someda, D. ; Tanaka, M.
|
||||
| 87.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Horita, Akihide
概要:
First, convergence properties in blind source separation (BSS) of convolutive mixtures are analyzed. A fully recurrent network is taken into account. Convergence is highly dependent on relation among signal source power,
…
transmission gain and delay in a mixing process. Especially, reverberations degrade separation performance. Second, a learning algorithm is proposed for this situation. In an unmixing block, feedback paths have an FIR filter. The filter coefficients are updated through the gradient algorithm starting from zero initial guess. The correction is exponentially scaled along the tap number. In other words, stepsize is exponentially weighted. Since the filter coefficients with a long delay are easily affected by the reverberations, their correction are suppressed. Exponential weighting is automatically adjusted by approximating an envelop of the filter coefficients in a learning process. Through simulation, good separation performance, which is the same as in no reverberations condition, can be achieved by the proposed method.
続きを見る
|
||||
| 88.
論文 |
論文
|
Nakayama, Kenji ; Ishikawa, Yutaka ; Kuraishi, Yoshiaki
|
||||
| 89.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Kashimoto, Hiroaki
概要:
金沢大学大学院自然科学研究科情報システム<br />This paper proposes a lattice predictor based adaptive Volterra filter (Lattice-AVF), and its
…
convergence property is analyzed. In the adaptive FIR Volterra filter (AVF), the eigen value spread of a correlation matrix is extremely amplified, and its convergence is very slow for gradient methods. A lattice predictor is employed for whitening the input signal. For stationary colored input signals, the Lattice-AVF can provide a fast convergence and the well reduced residual error. Its convergence is highly dependent on a time constant, used in updating the reflection coefficients. A very large time constant is required. In the case of nonstationary colored input signal, the eigen value spread after the Volterra polynomial is not so highly amplified. This means fast convergence will be expected, and effects of the whitening will be small. These properties are analyzed. A problem of asynchronous updating the reflection coefficients and the filter coefficients observed in linear lattice predictor based adaptive filters, is also observed in the LatticeAVF.
続きを見る
|
||||
| 90.
論文 |
論文
|
Tomikawa, K. ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />Recurrent Neural Network (RNN) can be applied to solve a combinatorial optimization problem. Howeve
…
r, the existence of the local minima in this model prevents its application to real world. In this paper, an analysis method for network dynamics based on eigenvalues and eigenvectors of a connection weight matrix is proposed. In this analysis, the transition of the number of negative eigenvalues and the movement of eigenspaces by increasing diagonal elements, which correspond to feedback loops of RNN, are discussed. From this analysis, it is confirmed that the number of negative eigenvalues decreases and the eigenspaces move toward the ascent side of the energy slope at the center point of the state space of RNN, as increasing diagonal elements. These behaviors of RNN contribute to the improvement for searching a solution of a combinatorial optimization problem. This analysis method is applied to a 2-dimensional example and 5 cities T.S. problems. From this analysis, it is theoretically found that the network performance of detecting the optimal solution can be improved by increasing the values of diagonal elements of connection matrix.
続きを見る
|
||||
| 91.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Nitta, M.
|
||||
| 92.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学大学院自然科学研究科情報システム<br />金沢大学工学部<br />A transfer function is constructed in a cascade form, using a low-order error-fre
…
e function and a high-order function. The high-order function is discretely optimized so that its error spectrum is suppressed by the error-free function. In order to save computing time, the error spectrum is equivalently evaluated in a time domain, and the coefficients are divided into small groups in a discrete optimization procedure
続きを見る
|
||||
| 93.
論文 |
論文
|
Nishiwaki, Takayuki ; Nakayama, Kenji ; Hirano, Akihiro
概要:
A network structure and its learning algorithm have been proposed for blind source separation applied to nonlinear mixtures. Nonlinearity is expressed by low-order polynomials, which are acceptable in many practical applications. A
…
separation block and a linearization block are cascaded. In the separation block, the cross terms are suppressed, and the signal sources are separated in each group, which include its high-order components. The high-order components are further suppressed through the linearization block. A learning algorithm minimizing the mutual information is applied to the separation block. A new learning algorithm is proposed for the linearization block. Simulation results, using 2-channel speech signals, instantaneous mixtures, and 2nd-order post nonlinear functions, show good separation performance.
続きを見る
|
||||
| 94.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakai, T.
|
||||
| 95.
論文 |
論文
|
95. On generalization of multilayer neural network applied to predicting protein secondary structure
Nakayama, Kenji ; Hirano, Akihiro ; Fukumura, K.
概要:
A learning process of a single neural network (SNN) to improve prediction accuracy of protein secondary structure is optimized. The protein secondary structures are predicted using a multiple alignment of amino acid as the input
…
data. A multi-modal neural network (MNN) has been proposed to improve the precision of prediction. This method uses five independent neural networks, and the final decision is made by averaging all outputs of five SNNs. In the proposed method, the same prediction accuracy can be achieved by using only a single NN and optimizing a learning process. In a learning process of protein structure prediction, over learning is easily occurred. So, the learning process is optimized so as to avoid the over learning. For this purpose, small learning rates, adding small random noise to the input data, and updating the connection weights by the average in some group are useful. The prediction accuracy 58% obtained by using the conventional SNN is improved to 66%, which is the same accuracy of the MNN, which needs five SNNs.
続きを見る
|
||||
| 96.
論文 |
論文
|
Nakayama, Kenji ; Inagaki, Kiyoto
|
||||
| 97.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakaguchi, H.
概要:
A sub-band adaptive filter, in which modulation and demodulation are employed, were proposed. The sampling rate can be minimized, while no aliasing occurs. In this paper, a polyphase and FFT realization of this sub-band adaptive filter is
…
proposed. In the polyphase and FFT realization for the transmultiplexer, the sampling rate reduction is the same as the number of the sub-bands. In the proposed method, however, they are different, so the conventional method cannot be applied. In the new realization, the polyphase filters are divided into a tapped delay line and an FFT part so as to be shared by all the sub-bands. Computational complexity is well reduced compared with the direct realization.
続きを見る
|
||||
| 98.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Someda, Daisuke ; Tanaka, Masahiko
概要:
金沢大学大学院自然科学研究科情報システム<br />This paper proposes an alternative learning algorithm for a stereophonic acoustic echo cancell
…
er without pre-processing which can identify the correct echo-paths. By dividing the filter coefficients into the former/latter parts and updating them alternatively, conditions both for unique solution and for perfect echo cancellation are satisfied. The learning for each part is switched from one part to the other when that part converges. Convergence analysis clarifies the condition for correct echo-path identification. For fast and stable convergence, a convergence detection and an adaptive step-size are introduced. The modification amount of the filter coefficients determines the convergence state and the step-size. Computer simulations show 10 dB smaller filter coefficient error than those of the conventional algorithms without pre-processing.
続きを見る
|
||||
| 99.
論文 |
論文
|
Yukawa, Akira ; Nakayama, Kenji ; Kawakami, Yuichi ; Hinooka, Kiyonobu ; Mizukami, Yoshihiko
|
||||
| 100.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakai, T.
概要:
金沢大学大学院自然科学研究科情報システム<br />In blind source separation, convergence and separation performances are highly dependent on a
…
relation between probability density functions (pdf) of signal sources and nonlinear functions used in updating coefficients of a separation block. This relation was analyzed based on kurtosis κ4. It was suggested that tanh y and y3, where y is the output, are useful nonlinear functions for super-Gaussian (κ4 > 0) and sub-Gaussian (κ4 < 0), respectively. In this paper, an adaptive nonlinear function is proposed. It has a form of f(y) = a tanh y + (1 - a)y3/4, where a is controlled by the kurtosis of the output signal yκ(n). It is assumed that the pdf p(y) of the output signal satisfies the stability condition f(y) = -(dp(y)/dy)/p(y). Based on this assumption, the parameter a and the kurtosis is related. This relation approximated by a function a = q(κ4). In a learning process, κ4(n) of the output signal is calculated at each sample n, and a(n) is determined by a(n) = q(κ4(n)). Then, the nonlinear function f(y) is adjusted. Blind separation of music signals of 2-5 channels were simulated. The proposed method is superior to a method, which switches tanh y and y3 based on polarity of κ4(n).
続きを見る
|
