※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
|
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
|
||||
| 4.
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
|
||||
| 5.
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
|
||||
| 6.
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
|
||||
| 7.
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
|
||||
| 8.
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
|
||||
| 9.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Ushimaru, S.
|
||||
| 10.
論文 |
論文
|
Wang, Youhua ; Terada, Yasuhiro ; Matsui, Minoru ; Iida, Kazuhiro ; Nakayama, Kenji
|
||||
| 11.
論文 |
論文
|
Nakayama, Kenji ; Katou, Haruo ; Hirano, Akihiro
|
||||
| 12.
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
|
||||
| 13.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakaguchi, Hiroaki
|
||||
| 14.
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
|
||||
| 15.
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
|
||||
| 16.
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
|
||||
| 17.
論文 |
論文
|
Tokui, N. ; Nakayama, Kenji ; Hirano, Akihiro
|
||||
| 18.
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
|
||||
| 19.
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
|
||||
| 20.
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
|
||||
| 21.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Horita, Akihide
|
||||
| 22.
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
|
||||
| 23.
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
|
||||
| 24.
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
|
||||
| 25.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Someda, D. ; Tanaka, M.
|
||||
| 26.
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
|
||||
| 27.
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
|
||||
| 28.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Nitta, M.
|
||||
| 29.
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
|
||||
| 30.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Sakai, T.
|
||||
| 31.
論文 |
論文
|
31. 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
|
||||
| 32.
論文 |
論文
|
Nakayama, Kenji ; Inagaki, Kiyoto
|
||||
| 33.
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
|
||||
| 34.
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
|
||||
| 35.
論文 |
論文
|
Tonomura, M. ; Nakayama, Kenji
概要:
The back-propagation algorithm is mainly used for multilayer perceptrons. This algorithm is, however, difficult to achieve high generalization when the number of training data is limited, that is sparse training data. In this paper, a new
…
learning algorithm is proposed. It combines the BP algorithm and modifies hyperplanes taking internal information into account. In other words, the hyperplanes are controlled by the distance between the hyperplanes and the critical training data, which locate close to the boundary. This algorithm works well for the sparse training data to achieve high generalization. In order to evaluate generalization, it is supposed that all data are normally distributed around the training data. Several simulations of pattern classification demonstrate efficiency of the proposed.
続きを見る
|
||||
| 36.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Kourin, Makoto
概要:
In this paper, a synthesis and learning method for the neural network with embedded gate units and a multi-dimensional input is proposed. When the input is multi-dimensional, gate functions are controlled in a multi-dimensional space.
…
In this case, a hypersurface, on which the gate function is formed should be optimized. Furthermore, the switching points should be considered on the unit input. An initialization and a control methods for gate functions, which optimize the hypersurface, the switching point and the inclination, are proposed. The stabilization methods, already proposed, are further modified to be applied to the multi-dimensional environment. The gate functions can be trained together with the connection weights. Discontinuous function approximation is demonstrated to confirm usefulness of the proposed method.
続きを見る
|
||||
| 37.
論文 |
論文
|
Nakayama, Kenji ; Katou, Haruo ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />In blind source separation (BSS) applications, the number of the signal sources is not known. When
…
the number of the sensors is less than that of the signal sources, this problem is called 'Over Complete BSS' (OC-BSS), which is a difficult problem due to lack of information in observations. In this paper, a feedback approach and its learning algorithm are proposed for the OC-BSS. The number of the outputs of an umixing block is set to be equal to that of the sensors. By assuming some condition, at least one output can separate a single signal source. This output is fed back to the inputs of the unmixing block, and is subtracted from the observations, in order to reduce the number of equivalent signal sources. Two kinds of feedback methods are proposed. One of them is direct subtraction and the other is sample elimination based on histogram of the feedback signal and the observed signals. The modified observations are further separated. The same process is repeated until all signal sources are separated. Performance of the proposed method is evaluated through computer simulation. The proposed method can improve a signal to interference ratio by the several dB compared to the conventional methods. © 2006 IEEE.
続きを見る
|
||||
| 38.
論文 |
論文
|
Boris, Jansen ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系<br />Over the years, many improvements and refinements of the backpropagation learning algorithm have b
…
een 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
|
||||
| 39.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />This paper presents implementations of an FIR adaptive filter with a large number of taps on nVIDIA
…
GeForce graphics processing unit (GPU) and CUOA software development environment. In order to overcome a long access latency for slow off-chip memory access, reduction of memory accesses by re-ordering and vector load/store operations and an increase of the number of threads are introduced. A tree adder is introduced to reduce the cost for summing thread outputs up. A simultaneous execution of multiple filters are also examined. On low-cost platform such as an Atom/ION nettop, GPU will accelerates the computation by almost three times. For simultaneous multiple simulations such as an ensemble averaging, a GPU with a large number of processing elements outperforms a dual-core CPU; almost six times faster for 16 runs. © 2010 IEEE
続きを見る
|
||||
| 40.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />FeedForward (FF-) Blind Source Separation (BSS) systems have some degree of freedom in the solutio
…
n space. Therefore, signal distortion is likely to occur. First, a criterion for the signal distortion is discussed. Properties of conventional methods proposed to suppress the signal distortion are analyzed. Next, a general condition for complete separation and distortion-free is derived for multi-channel FF-BSS systems. This condition is incorporated in learning algorithms as a distortion-free constraint. Computer simulations using speech signals and stationary colored signals are performed for the conventional methods and for the new learning algorithms employing the proposed distortion-free constraint. The proposed method can well suppress signal distortion, while maintaining a high source separation performance.
続きを見る
|
||||
| 41.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系
|
||||
| 42.
論文 |
論文
|
Nakayama, Kenji ; Kaneda, Yasuaki ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系
|
||||
| 43.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />出版者許諾要件を調査中.
|
||||
| 44.
論文 |
論文
|
Katou, Haruo ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />本稿では,信号源の数がセンサ数より多いオーバーコンプリート・ブラインド信号源分離においてフィードバック形構成法と分離回路の学習アルゴリズムを提案する.まず,1巡目で信号源の分離を行う.ここでは,一
…
つの信号源が複数の出力に含まれないことを分離の条件とする.このための学習法を提案している.センサ数を信号源数の約半分以上とすることにより,1巡目の信号源分離で少なくとも1個の出力に単一信号源を分離できる.この出力を単一信号源の特徴を利用して検出する.更に,この出力をフィードバックして観測信号からキャンセルすることにより,等価的に信号源の数を低減する.当該出力と混合過程の情報を使ってキャンセルする他,観測信号と当該出力のヒストグラムを使ってキャンセルする方法を組み合わせることにより,条件不足の問題を解消する.2巡目では,観測信号に含まれる信号源が1個少ない状態で1巡目と同じ処理を行う.このように,提案法では,観測信号における信号源の数を1個ずつ減らしながら信号源分離を繰り返す.信号源として音声を用いたシミュレーションにより,従来法との比較を行い,提案方法の有効性を確認している. A feedback structure and its learning algorithm are proposed for overcomplete blind source separation, where the number of sources is larger than that of sensors. In the first phase, the signal sources are separated so as to satisfy the condition, under which one signal source is not included in different outputs. By setting the number of the sensors to be more than a half of the sources, at least one output includes a single source. This output is selected based on property of a single source. Furthermore, this output is fed back and cancelled from observed signals. The selected single source, information about a mixing process and histogram of the observations and the separated source are used for this cancellation. In the second phase, the same process is carried out by using the modified observations, in which the source, separated in the first phase, is cancelled. Like this, in the proposed method, the source separation is repeated by reducing the number of equivalent sources. Simulations using speech signals demonstrate usefulness of the proposed method compared to the conventional methods.
続きを見る
|
||||
| 45.
論文 |
論文
|
Nakayama, Kenji ; Katou, Haruo ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系
|
||||
| 46.
論文 |
論文
|
Nakayama, Kenji ; Kaneda, Yasuaki ; Hirano, Akihiro ; Haruta, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />Multilayer neural networks (MLNN) and the FFT amplitude of brain waves have been applied to dasiaB
…
rain Computer Interfacepsila (BCI). In this paper, a magnetoencephalograph (MEG) system, dasiaMEGvisionpsila developed by Yokogawa Corporation, is used to measure brain activities. MEGvision is a 160-channel whole-head MEG system. Channels are selected from 8 main regions, a frontal lobe, a temporal lobe, a parietal lobe and a occipital lobe, located in the left and the right sides of the brain. The 8 channels, located at the central point in the 8 lobes, are initially selected. Optimum channels are searched for in the same lobe as the initial channels in order to achieve high classification accuracy. Two subjects and four mental tasks, including relaxed situation, multiplication, playing sport and rotating an object, are used. The brain waves are measured 10 times for one subject and one mental task. Among them, 8 data sets are used for training the MLNN, and the remaining 2 data sets are used for testing. 5 kinds of combinations of 2 data sets are selected for testing. Rates of correct classification by using the initial channels are 82:5 ~ 90%. By optimizing the channels, the accuracy is improved up to 85:0 ~97:5%, which is very high accuracy. Furthermore, contributions of the brain waves in the 8 lobes are analyzed.
続きを見る
|
||||
| 47.
論文 |
論文
|
Rebolledo Méndez, Jovan David ; Nakayama, Kenji ; Méndez Morales, Adolfo
概要:
金沢大学理工研究域 電子情報学系<br />Visually impaired people face many obstacles to interact properly with the current screen / visual
…
ly oriented technologies, as monitors or in using Internet, being destined just for users with good visual ability. This paper refers to a specification of a proposal of an innovative way to safe that obstacle for visually impaired people and provide them a device that will make them understand the normal output from a computer. This method attempts to use the basic element: a wind element (wixel). It will be explained its technical specification, methodology, materials, usages, benefits, current development and future work as well.
続きを見る
|
||||
| 48.
論文 |
論文
|
Nakayama, Kenji ; Inagaki, Kiyoto
概要:
金沢大学理工研究域 電子情報学系<br />Brain Computer Interface (BCI) is one of hopeful interface technologies between human and machine.
…
However, brain waves are very weak and there exist many kinds of noises. Therefore, what kinds of features are useful, how to extract the useful features, how to suppress noises, and so on are very important. On the other hand, neural networks are very useful technology for pattern classification. Especially, multilayer neural networks trained through the error back-propagation algorithm have been widely used in a wide variety of field. In this paper, the neural network is applied to the BCI. Amplitude of the FFT of the brain waves are used for the input data. Several kinds of techniques are introduced in this paper. Segmentation along the time axis for fast response, nonlinear normalization for emphasizing important information with small magnitude, averaging samples of the brain waves for suppressing noise effects and reduction in the number of the samples for achieving a small size network, and so on are newly introduced. Simulation was carried out by using the brain waves, which are available from the web site of Colorado state university. The number of mental tasks is five. Ten data sets for each mental task are prepared. Among them, 9 data sets are used for training, and the rest one data set is used for testing. Selection of the one data set for testing is changed and accuracy of the correct classifications are averaged over the possible selections. Approximately, 80% of correct classification of the brain waves is obtained, which is higher than the conventional. © 2006 IEEE.
続きを見る
|
||||
| 49.
論文 |
論文
|
Kaneda, Yasuaki ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />脳波のFFTと階層形ニューラルネットワークを用いるブレイン・コンピュータ・インタフェイス(BCI)に関して,以前に前処理の方法をいくつか提案し,メンタルタスクの分類性能を向上した.本稿では,まず,
…
階層形ニューラルネットワークでメンタルタスクを分類するために用いられる特徴の解析を行った.特徴は結合荷重の分布に基づいて解析した.隠れ層から出力層への結合荷重はメンタルタスクに対して独立になる傾向があった.従って,入力層から各メンタルタスクに対応する隠れユニットへの結合荷重分布がメンタルタスク毎の特徴を表している.次に,汎化能力を向上する2通りの学習法について検討を行った.一つは,ニューラルネットワークの入力データに乱数を加える方法であり,もう一つは,結合荷重を圧縮する方法する方法である.シミュレーションの結果,いずれの方法もテストデータに対する分類性能を向上することが出来たが,乱数を加える方法が有効であることが分かった. In this paper, a multilayer neural network is applied to 'Brain Computer Interface' (BCI), which is one of hopeful interface technologies between humans and machines. Amplitude of the FFT of the brain waves are used for the input data. Several techniques have been introduced for pre-processing the brain waves. They include segmentation along the time axis for fast response, nonlinear normalization to emphasize important information, averaging samples of the brain waves to suppress noise effects, reduction in the number of the samples to realize a small size network, and so on. In this paper, two kinds of generalization techniques, including adding small random noises to the input data and decaying connection weight magnitude, are applied. Their usefulness are analyzed and compared base on correct and error classifications. Simulation is carried out by using the brain waves, which are available from the web site of Colorado State University. The number of mental tasks is five. Some data sets are used for training the multilayer neural network, and the remaining data sets are used for testing. In our previous work, classification accuracy of 64%〜74% for the test data have been achieved. In this paper, by applying the generalization techniques, the accuracy can be improved up to 80%～88%.
続きを見る
|
||||
| 50.
論文 |
論文
|
Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系
|
||||
| 51.
論文 |
論文
|
Rebolledo Méndez, Jovan David ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系<br />A Natural Language Generation (NLG) engine is proposed based on the combination of NLG and Expert
…
Systems. The combination of these techniques paves the way to employ user defined behaviors in virtual worlds as inputs to an expert system. Adaptive Algorithms can then be used to retrieve information from the Internet to give feedback to the user via the NLG engine. The combination of these AI techniques can bring about some benefits such as believability in the interaction between AI-driven and human-driven avatars in virtual worlds. © 2008 Springer-Verlag Berlin Heidelberg.
続きを見る
|
||||
| 52.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />This paper presents an implementation of a stereophonic acoustic echo canceller on nVIDIA GeForce g
…
raphics processor and CUDA software development environment. For ef.ciency, fast shared memory has been used as much as possilbe. A tree adder is introduced to reduce the cost for summing thread outputs up. The performance evaluation results suggest that Even a low-cost GPU's with a small number of shader processor greatly helps the echo cancellation for low-cost PC-based teleconferencing. ©2009 IEEE..
続きを見る
|
||||
| 53.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />In Blind Source Separation (BSS), a eparation block is trained so as to make the output signals to
…
be statistically independent. In this case, the independency is able to be increased by changing frequency response of the output signals, resulting in signal distortion. Especially, a feed-forward BSS (FF-BSS) has some degree of freedom in the separation block, and the signal distortion will be caused. The signal distortion is evaluated as difference between the output signal and the signal source in the measured signal. Some equations are derived from the conditions of complete separation and signal distortion free. They are used as the distortion free constraint in the conventional learning process [11]. On the other hand, a feedback BSS (FB-BSS) has a solution, which can satisfy both complete separation and distortion free. In this paper, the learning algorithm with the distortion free constraint is applied to the FF-BSS in time domain. Many kinds of signal sources are used in simulation in order to compare the proposed method and the conventional, in which difference between the output signals and the measured signals is included in the cost function [4]. Furthermore, the FB-BSS is also evaluated.ブラインド信号源分離では(BSS) は分離回路がその出力信号が統計的に独立になるように学習される．この場合，出力信号の周波数特性が変化することにより，独立性が高まることもあるので，信号歪みが生じる可能性がある．特に，フィードフォワード形BSS(FF-BSS)は分離回路における自由度が高く，信号歪みを生じる可能性がある．信号歪みの基準を観測信号に含まれる信号源と考え，完全分離の条件と信号無歪の条件から導かれた制約条件を学習に加味する信号歪み抑制学習法を提案した[11]．信号源をsi，観測信号をxi，出力信号をyi とするとき，信号を分離するとともにyi をxi におけるsi 成分に近づけることができる．これに対し，観測信号と出力信号の差を評価関数に追加する従来法では，観測信号に含まれる複数の信号源の影響で信号源分離が充分ではない．一方，フィードバック形BSS(FBBSS)では，信号源分離と信号歪み抑制の条件を同時に満たす回が存在する．本稿では信号歪み抑制学習法を時間領域で学習するFF-BSS に適用し，種々の信号源を 使って従来方式[4] と比較することによりその特性を解 析する．同時に，FB-BSS の有効性も検証する．
続きを見る
|
||||
| 54.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系
|
||||
| 55.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系<br />Over the years, many improvements and refinements to the backpropagation learning algorithm have b
…
een 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
|
||||
| 56.
論文 |
論文
|
Nakayama, Kenji ; Higashi, Shoya ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />A noise spectral estimation method, which is used in spectral suppression noise cancellers, is pro
…
posed for highly non-stationary noise environments. Speech and non-speech frames are detected by using the entropy-based voice activity detector (VAD). An adaptive normalization parameter and a variable threshold are newly introduced for the VAD. They are very useful for rapid change in the noise spectrum and power. Furthermore, a recursive averaging method is applied to estimating the noise spectrum in the non-speech frames. In this method, an adaptive smoothing parameter is proposed, based on speech presence probability. Simulations are carried out by using many kinds of noises, including white, babble, car, pink, factory and tank, which are changed from one to the other. The segmental SNR is improved by 2.0 ~3.8dB, and noise spectral estimation error is improved by 3.2 ~ 4.7dB for the white noise and the babble noise, which are changed from one to the other.
続きを見る
|
||||
| 57.
論文 |
論文
|
Nakayama, Kenji ; Suzuki, H. ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />In this paper, new approaches to noise spectrum estimation and spectral gain control are proposed
…
for noise spectral suppressors. First, the speech absent frames are detected by using spectral entropy. In the speech absent frames, a weighting factor used in estimating the noise spectrum is modified so as to emphasize effect of the noisy speech signal. Next, a spectral gain is more reduced by multiplying a factor in order to suppress effects of the noise in the speech absent frames. Furthermore, in the speech present frames, in order to reduce signal distortion, the spectral gain is controlled to be unity based on an SNR calculated by using a ridgeline spectrum. Finally, the original noisy speech is added to the estimated speech in some ratio. This ratio is controlled by the long term averaged SNR of the estimated noise and the noisy speech. Computer simulations by using speech signals, the white noise, the car noise and the bubble noise, which are available in public, have been carried out for the conventional methods and the proposed method. The proposed method can improve a segmental SNR and speech quality compared to the conventional methods. Especially, it is useful for the bubble noise. ©2007 IEEE.
続きを見る
|
||||
| 58.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系<br />階層型ニュートラルネットワークの学習法としてよく用いられるバックプロパゲーション(BP)アルゴリズムに対して，学習の収束性を改善する多くの方法が提案されている. 本稿では，BPアルゴリズムやBPR
…
OP法などの類似する学習法を対象として，新しい適応形ペナルティに基づく学習法を提案する.学習法で用いられる出力誤差をペナルティにより増減する.ペナルティは出力が目標値と同じ放物線上にあれば小さく，そうでなければ大きく制御される. これにより，局所解に陥ることを防ぐことが出来，最適解への収束性を高めることが出来る. 多くの例を用いてシュミュレーションを行った結果，提案方法の有効性が確認できた.
続きを見る
|
||||
| 59.
論文 |
論文
|
Nakayama, Kenji ; Hirano, Akihiro ; Kashimoto, Hiroaki
概要:
金沢大学理工研究域 電子情報学系<br />適応Volterraフィルタ(AVF)は一般的な非線形特性を表現できるが, 収束が遅いという問題がある.これに対して, 入力信号を白色化する方法が提案されている.その中で, ラチス形予測器を用いる
…
方法が収束性の点で優れている.しかし, 反射係数とフィルタ係数の更新における非同期の問題がある.線形AVFではこの問題を解決する同期形学習法が提案され, 収束特性を大幅に改善している.本稿では, ラチス形AVFにおける非線形部に対する同期学習法を新たに提案する.シミュレーションにより, 非定常有色信号に対して有効であることを確認した. The adaptive Volterra filters (AVF) can represent general nonlinearity. However, its convergence speed is very slow. In order to solve this problem, techniques of whitening the input signal have been proposed. Among them, the lattice prediction error filter is useful. The lattice predictor based AVF (Lattice-AVF) has some problem. Updating the reflection coefficients and the filter coefficients is asynchronous, causing poor convergence properties. In order to solve this problem, the synchronized learning algorithm has been proposed for the linear adaptive filters. In this paper, a new synchronized learning algorithm is proposed for the nonlinear part of the Lattice-AVF. Its usefulness is confirmed though simulation using nonstationary colored signals.
続きを見る
|
||||
| 60.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji
概要:
金沢大学理工研究域 電子情報学系<br />This paper presents an computationally ef cient implementation of sparse-tap FIR adaptive lters wi
…
th tapposition control on Intel IA-32 processors with single-instruction multiple-data (SIMD) capability. In order to overcome randomorder memory access which prevents a ectorization, a blockbased processing and a re-ordering buffer are introduced. A dynamic register allocation and the use of memory-to-register operations help the maximization of the loop-unrolling level. Up to 66percent speedup is achieved.<br />Organized by the Electrical Engineering/Electronics, Computer, Telecommunications, and Information Technology Association (ECTI) Co-organized by GCEO-NGIT, Hokkaido University Technical sponsored by IEEE Circuits and Systems Society In cooperation with the Institute of Electronics, Information and Communication Engineering (IEICE)
続きを見る
|
||||
| 61.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />Source separation and signal distortion in three kinds of BSSs with convolutive mixture are analyz
…
ed. They include a feedforward 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
|
||||
| 62.
論文 |
論文
|
Nakayama, Kenji ; Horita, Hiroki ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />FFT and Multilayer neural networks (MLNN) have been applied to 'Brain Computer Interface' (BCI). I
…
n this paper, in order to extract features of mental tasks, individual feature of brain waves of each channel is emphasized. Since the brain wave in some interval can be regarded as a vector, Gram-Schmidt orthogonalization is applied for this purpose. There exists degree of freedom in the channel order to be orthogonalized. Effect of the channel order on classification accuracy is investigated. Next, two channel orders are used for generating the MLNN input data. Two kinds of methods using a single NN and double NNs are examined. Furthermore, a generalization method, adding small random numbers to the MLNN input data, is applied. Simulations are carried out by using the brain waves, available from the Colorado State University website. By using the orthogonal components, a correct classification rate P c can be improved from 70% to 78%, an incorrect classification rate P e can be suppressed from 10% to 8%. As a result, a rate R c ∈=∈P c /(P c ∈+∈P e ) can be improved from 0.875 to 0.907. When two different channel orders are used, P e can be drastically suppressed from 10% to 2%, and R c can be improved up to 0.973. The generalization method is useful especially for using a sigle channel order. P c can be increased up to 84~88% and P e can be suppressed down to 2~4%, resulting in R c ∈=∈0.957~0.977. © 2008 Springer-Verlag Berlin Heidelberg.
続きを見る
|
||||
| 63.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />Source separation and signal distortion are theoretically analyzed for the FF-BSS systems implemen
…
ted 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
|
||||
| 64.
論文 |
論文
|
Rebolledo Méndez, Jovan David ; Nakayama, Kenji
概要:
金沢大学理工研究域電子情報学系<br />A Natural Language Generation (NLG) engine is proposed based on the combination of NLG and Expert S
…
ystems. The combination of these techniques paves the way to employ user defined behaviors in virtual worlds as inputs to an expert system. Adaptive Algorithms can then be used to retrieve information from the Internet to give feedback to the user via the NLG engine. The combination of these AI techniques can bring about some benefits such as believability in the interaction between AI-driven and human-driven avatars in virtual worlds. © 2008 Springer-Verlag Berlin Heidelberg.
続きを見る
|
||||
| 65.
論文 |
論文
|
Nakayama, Kenji ; Inagaki, Kiyoto
概要:
Brain Computer Interface (BCI) is one of hopeful interface technologies between human and machine. However, brain waves
…
are very weak and there exist many kinds of noises. Therefore, what kinds of features are useful, how to extract the useful features, how to suppress noises, and so on are very important. On the other hand, neural networks are very useful technology for pattern classification. Especially, multilayer neural networks trained through the error back-propagation algorithm have been widely used in a wide variety of field. In this paper, the neural network is applied to the BCI. Amplitude of the FFT of the brain waves are used for the input data. Several kinds of techniques are introduced in this paper. Segmentation along the time axis for fast response, nonlinear normalization for emphasizing important information with small magnitude, averaging samples of the brain waves for suppressing noise effects and reduction in the number of the samples for achieving a small size network, and so on are newly introduced. Simulation was carried out by using the brain waves, which are available from the web site of Colorado state university. The number of mental tasks is five. Ten data sets for each mental task are prepared. Among them, 9 data sets are used for training, and the rest one data set is used for testing. Selection of the one data set for testing is changed and accuracy of the correct classifications are averaged over the possible selections. Approximately, 80% of correct classification of the brain waves is obtained, which is higher than the conventional. © 2006 IEEE.
続きを見る
|
||||
| 66.
論文 |
論文
|
Ritthipravat, P. ; Nakayama, Kenji
|
||||
| 67.
論文 |
論文
|
Nakayama, Kenji ; Horita, Akihide ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />Feed-Forward (FF-) and Feed-Back (FB-) structures have been proposed for Blind Source Separation (
…
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
|
||||
| 68.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />Source separation and signal distortion are theoretically analyzed for the FF-BSS systems implemen
…
ted 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
|
||||
| 69.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Arai, Shoji
|
||||
| 70.
論文 |
論文
|
Nakayama, Kenji ; Higashi, Shoya ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系<br />A noise spectral estimation method, which is used in spectral suppression noise cancellers, is pro
…
posed for highly non-stationary noise environments. Speech and non-speech frames are detected by using the entropy-based voice activity detector (VAD). An adaptive normalization parameter and a variable threshold are newly introduced for the VAD. They are very useful for rapid change in the noise spectrum and power. Furthermore, a recursive averaging method is applied to estimating the noise spectrum in the non-speech frames. In this method, an adaptive smoothing parameter is proposed, based on speech presence probability. Simulations are carried out by using many kinds of noises, including white, babble, car, pink, factory and tank, which are changed from one to the other. The segmental SNR is improved by 2.0 ~3.8dB, and noise spectral estimation error is improved by 3.2 ~ 4.7dB for the white noise and the babble noise, which are changed from one to the other.
続きを見る
|
||||
| 71.
論文 |
論文
|
Nakayama, Kenji ; Horita, Akihide ; Hirano, Akihiro
概要:
金沢大学理工研究域 電子情報学系
|
||||
| 72.
論文 |
論文
|
Jansen, Boris ; Nakayama, Kenji
概要:
Over the years, many improvements and refinements of the backpropagation learning algorithm have 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
|
||||
| 73.
論文 |
論文
|
Horita, Akihide ; Nakayama, Kenji ; Hirano, Akihiro ; Dejima, Yasuhiro
概要:
金沢大学理工研究域 電子情報学系<br />畳み込み混合過程におけるフィードフォワード(FF-)形ブラインドソースセパレーション(BSS)では自由度が存在するため信号歪みが生じる.我々は,信号源-センサーが2チャンネルの場合において,完全分
…
離と信号無歪みの条件を制約条件として課す信号歪み抑制学習アルゴリズムを時間領域,周波数領域のFF-BSSに対して提案してきた.本稿では,信号歪み抑制の制約条件を多チャンネルに拡張し,かつ,計算の複雑さを軽減するために制約条件を近似する方式を提案する.音声を用いたコンピュータシミュレーションによってその近似制約方式と厳密制約方式がほぼ同等の分離性能と信号歪み抑制が得られることを確認した.また,3チャンネルにおいても,従来方式より特性が改善されることを確認した. Feed-forward Blind Source Separation (FF-BSS) systems have some degree of freedom in the solution space, and signal distortion is likely to occur in convolutive mixtures. Previously, a condition for complete separation and distortion free has been derived for 2-channel FF-BSS. This condition has been applied to the learning algorithms as a distortion free constraint in both the time and frequency domains. In this paper, the condition is further extended to multiple channel FF-BSSs. This condition requires the a high computational complexity to be applied to the learning process as a constraint. An approximate constraint is proposed in order to relax the high computational load. In comparison with the original constraint, computer simulations have demonstrated that the approximation can obtain similar performances with respect to source separation as well as signal distortion using speech signals. Furthermore, the performances can be improved compared to the conventionals for three channels.
続きを見る
|
||||
| 74.
論文 |
論文
|
Hirano, Akihiro ; Nakayama, Kenji ; Takebe, Kazue
|
||||
| 75.
論文 |
論文
|
中山, 謙二 ; Nakayama, Kenji
概要:
ブレインコンピュータインタフェイス(BCI)において,多チャネルで測定された脳波をチャネル間で直交化することにより脳波の特徴を強調する方法を提案した.並列構成された複数の階層形ニューラルネットワーク(MLNN)を用いてメンタルタスク(MT)
…
を分類し,それらの結果を統合する方法により, MTの分類性能が大幅に向上した.他の方式として,脳波の特徴を強調する部分空間フィルタと複数の2分類器,その出力を誤差訂正符号化する方式を提案し,高い分類性能を得た.<br />Brain Computer Interface(BCI) system has been developed. A method to emphasize features of the brain waves has been proposed. The orthogonalized components and parallel multi-layer neural networks are used to classify themental tasks. These results are averaged to obtain the final result. The high performance for mental task classification has been obtained. Another method, combining the special filter, binary classifiers and error correcting code, has beenproposed, resulting in high performance.<br />研究課題/領域番号:21560393, 研究期間(年度):2009-2011
続きを見る
|
||||
| 76.
論文 |
論文
|
中山, 謙二 ; Nakayama, Kenji
概要:
1)オーバーコンプリート形ブラインド信号源分離(OC-BSS)平成17年度に提案した,フィードバック形BSSにおいて,フィードバックにより信号源を相殺するための好しい学習法を提案した。混合過程に関する情報を使う方法と信号のヒストグラムを使う
…
2つの方法を提案した。さらに,ヒストグラムを使う方法に関して,信号歪みを低減する方法を提案した。信号歪みを雑音と見なして,スペクトルサプレッション法により雑音スペクトルを抑制する方法である。音声の信号源を3個,センサーを2個としたときのシミュレーションを行い,従来法に比べて,信号対干渉比が約10dB改善された。2)非線形混合過程に対するブラインド信号源分離信号源のグループ分離と線形化を縦続接続する方式を以前に提案したが,その学習法に関して改良を行った。特に,線形化に関して,「初期値の設定法」及び「パラメータの学習法」に関して新しい方法を提案し,信号源の分離特性を大幅に改善した。また,信号源とセンサーの位置関係と必要なセンサー数の関係についても解析し,実用化における指針を与えた。3)ブラインド信号源分離における信号歪みの低減フィードフォワード形(FF-)BSSに対して,信号歪みを抑制する新しい学習法を提案した。従来の学習法に信号歪みを抑制する制約条件を課す方法である。2チャネルと3チャネルについてシミュレーションを行い,分離特性と信号歪みを評価した。その結果,従来の信号歪み抑制形学習法に比べて大幅な特性改善を実現した。さらに,フィードバック形(FB-)BSSとFF-BSSが各々有効に使用できる条件を明らかにした。<br />A feedback approach and its learning algorithm are proposed for the OC-BSS. By using the sensors more than a half of the sources, at least one output can separate a single signal source. This output is fed back to the inputs of the separation block, and is subtracted from the observations, in order to reduce the number of equivalent signal sources. Two kinds of feedback methods are proposed, which are direct subtraction and sample elimination based on histogram of the observed signals and the separated signal above. In the latter process, signal distortion is further suppressed by the spectral suppression technique. The proposed method can improve a signal to interference ratio by 6〜10 dB compared to the conventional methods.Source separation and signal distortion are theoretically analyzed in blind source separation (BSS) systems implemented in both the time and the frequency domains. Feedforward (FF-) BSS systems have some degree of freedom in the solution space. Therefore, signal distortion is likely to occur. Next, a condition for complete separation and distortion free is derived for multi-channel FF-BSS systems. This condition is incorporated in learning algorithms as a distortion free constraint. Computer simulations using speech signals and stationary colored signals are performed for conventional methods and the new learning algorithms employing the proposed distortion free constraint. The proposed method can drastically suppress signal distortion, while maintaining a high separation performance.<br />研究課題/領域番号:17560335, 研究期間(年度):2005-2006<br />出典：「非線形畳み込み混合過程におけるオーバーコンプリート形ブラインド信号源分離」研究成果報告書 課題番号17560335 (KAKEN：科学研究費助成事業データベース（国立情報学研究所）) 本文データは著者版報告書より作成
続きを見る
|
||||
| 77.
論文 |
論文
|
中山, 謙二 ; Nakayama, Kenji
概要:
金沢大学自然科学研究科<br />信号源やその伝達・混合過程に関する事前情報がなく,それらの統計的な性質のみを仮定して信号源を分離するブラインド形信号源分離(BSS)に関して研究を行った.特に,伝達・混合過程に非線形を含む場合は信号源の混合
…
も複雑になり,それらを分離することは一般に難しい.本研究では,信号源の高次項を含む信号群を分離する過程と,その後に高次項を抑制する線形化の過程を縦続接続する方法を提案した.また,この構成法に対する学習法を新たに提案した.音声の信号源が2個の場合と3個の場合についてシミュレーションを行った.観測信号に比べて,干渉成分と非線形成分(高次項)は約20dB程度減衰しており,有効性が確認できた.本方法では,信号源の数より多い観測センサーが必要とされる.これに関して,非線形成分の大きさと必要とされる観測センサー数の関係について解析し,実際の応用場面では,センサー数を低減できることを示した.さらに,線形化→信号群分離→線形化というサンドウィッチ構造を提案し,少ないセンサー数で良好な分離特性を得た.ブラインド形信号源分離では,分離回路において信号が歪むという問題がある.これに関して,フィードフォワード形BSS(FF-BSS)とフィードバック形BSS(FB-BSS)について解析を行い,信号歪みのメカニズムを始めて明らかにした.FF-BSSは分離回路の自由度が高く,信号歪みが生じる.一方,FB-BSSは観測信号から信号歪みを生じることなく,分離が可能であることが明らかになった.さらに,本研究では,FF-BSSにおいても信号歪みを抑制するための制約条件を付加した学習法を提案し,時間領域における学習で従来法に比べて信号歪みを抑えることができた.以上,本研究では,非線形混合過程におけるBSSに対して新しい方法を提案し,さらに,従来行われていなかった信号歪みの問題を解析し,新しい学習法を提案した.これらにより,BSSをより現実的な場面に応用することが可能になり,また,分離後においても良好な音質が保証される.<br />In practical applications of BSS, processes of generating mixing and sensing signals include nonlinearity, caused by loud speakers, microphones, amplifiers and so on. BSS, cascading a signal group separation block and a linearization block has been proposed for low-order nonlinear mixtures. In the separation block, the signal sources are separated into each group, including its high-order components. The high-order components are further suppressed through the linearization block.In this report, separation performance of the nonlinear BSS is analyzed from several view points. The number of the sensors is increased from that of the signal sources in order to cancel the interference. Moreover, the interference components is decided by a ratio of the nonlinear and the linear components. A relation between the ratio of the components and the number of the sensors is analyzed. The number of the sensors can be reduced when the ratio of the nonlinearity is small. And a Cascade Form BSS Connecting Linearization and Source Separation and Linearization is analyzed.Next, effects of the initial guess of the separation matrix is analyzed. The training was carried out using 50 independent random initial guess, and good separation is obtained by a 25% probability. Moreover, effect of including 3rd-order terms is analyzed. When the 3rd-order term is under 10%, good separation performance can be obtained.<br />研究課題/領域番号:15560323, 研究期間(年度):2003 – 2004<br />出典：「非線形たたみ込み混合過程におけるブラインド形信号源分離・推定法」研究成果報告書 課題番号15560323（KAKEN：科学研究費助成事業データベース（国立情報学研究所））（https://kaken.nii.ac.jp/ja/report/KAKENHI-PROJECT-15560323/155603232004kenkyu_seika_hokoku_gaiyo/)を加工して作成
続きを見る
|
||||
| 78.
論文 |
論文
|
中山, 謙二 ; Nakayama, Kenji
概要:
金沢大学自然科学研究科<br />自然現象や社会現象等の時間的な変化を予測することは自然破壊,環境問題,人口問題,経済危機等に対応するために非常に重要である。このような現象の多くは非線形であり,非線形予測器の研究も多く行われている。本研究で
…
は,非線形予測を過去のサンプルから未来のサンプルへの写像としてとらえ,非線形なパターン写像の有効な手法であるニューラルネットワークと線形予測を組み合わせたハイブリッド形非線形予測器を開発した。(1)非線形予測器の提案階層形ニューラルネットワークによる非線形予測を入力側に,FIRフィルタによる線形予測を出力側に配置した縦続形のハイブリッド形予測器を提案した。階層形ニューラルネットワークの出力ユニットは線形素子であり,線形予測の能力も有している。(2)学習方法の提案非線形予測器と線形予測器をこの順に分離学習する方法を提案した。次に,雑音を含む非線形時系列の予測において,強化学習法を提案した。誤差逆伝播学習法の過程に結合荷重の強化を組み込んだ方法である。ニューロンの入力値が活性化関数の飽和領域にシフトされ雑音の影響が低減される。(3)時系列データの非線形性を解析する方法の提案類似する過去のサンプルから予測されるサンプルの分散で非線形性を評価する方法を提案した。実際の非線形データ及び線形システムから生成した時系列を用いてその有効性を確認した。これに基づいて,非線形予測に必要なニューラルネットワークの規模を推定する方法を提案し,最小構成を可能とした。(4)実際の非線形時系列の予測太陽黒点,湖の水位,カオス,及び霧発生の時間的な変化の予測に本予測器を適用し,その有効性を調べた結果,従来方法に比べて本方式が常に最小の予測誤差を与えていることを確認した。<br />Now a day, we have a lot of problems, environmental disruption, environmental pollution, economic crisis, population problem, natural disaster, nature conservation, and so on. In order to solve these problems, it is very important to analyze progress of these phenomena. These phenomena can be regarded as time series. Mainly they are nonlinear time series. So, nonlinear prediction becomes very important.(1) A Nonlinear PredictorIn this research project, we have developed a hybrid nonlinear predictor, which combines a neural network and a feed-forward linear predictor. Since the neural network has linear output unit, most of nonlinear part and some linear part can be predicted by the neural network. The remaining part is predicted by the linear predictor.(2) Learning AlgorithmsAn improved learning algorithm has been proposed, which separately optimize the neural network and the linear predictor in this order. An enhanced learning algorithm has been proposed for noisy nonlinear time series prediction.(3) Nonlinearity Analysis of Time SeriesPrediction is the mapping from the past sample x(n-1)=[x(n-1),x(n-2),..,x(n-N)] to the next sample x(n). When the past samples x(nィイD21ィエD2-1) and x(nィイD22ィエD2-1) are similar, however, the next samples x(nィイD21ィエD2) and x(nィイD22ィエD2) are far from to each other, then, nonlinearity of this time series is high. A measure, which can evaluate this property has been introduced.(4) Prediction of Real Nonlinear Time SeriesThe proposed method was applied to many the real nonlinear time series, including Chaos, water levels of some lake, fog generation, and so on. The proposed hybrid nonlinear predictor demonstrated good performance compared with the conventional methods.<br />研究課題/領域番号:10650039, 研究期間(年度):1998 – 1999<br />出典：「ハイブリッド型非線形予測器の最小構成法と学習アルゴリズム」研究成果報告書 課題番号10650039（KAKEN：科学研究費助成事業データベース（国立情報学研究所）） （https://kaken.nii.ac.jp/ja/report/KAKENHI-PROJECT-10650357/106503571999kenkyu_seika_hokoku_gaiyo/）を加工して作成
続きを見る
|
