Timm HOHR Andreas SCHENK Andreas WETTSTEIN Wolfgang FICHTNER
The density gradient (DG) model is tested for its ability to describe tunneling currents through thin insulating barriers. Simulations of single barriers (MOS diodes, MOSFETs) and double barriers (RTDs) show the limitations of the DG model. For comparison, direct tunneling currents are calculated with the Schrodinger-Bardeen method and used as benchmark. The negative differential resistance (NDR) observed in simulating tunneling currents with the DG model turns out to be an artifact related to large density differences in the semiconductor regions. Such spurious NDR occurs both for single and double barriers and vanishes, if all semiconductor regions are equally doped.
Seungjin CHOI Andrzej CICHOCKI Liqing ZHANG Shun-ichi AMARI
This paper addresses a maximum likelihood method for source separation in the case of overdetermined mixtures corrupted by additive white Gaussian noise. We consider an approximate likelihood which is based on the Laplace approximation and develop a natural gradient adaptation algorithm to find a local maximum of the corresponding approximate likelihood. We present a detailed mathematical derivation of the algorithm using the Lie group invariance. Useful behavior of the algorithm is verified by numerical experiments.
Zhe-Ming LU Bian YANG Sheng-He SUN
Vector quantization (VQ) is an attractive image compression technique. VQ utilizes the high correlation between neighboring pixels in a block, but disregards the high correlation between the adjacent blocks. Unlike VQ, side-match VQ (SMVQ) exploits codeword information of two encoded adjacent blocks, the upper and left blocks, to encode the current input vector. However, SMVQ is a fixed bit rate compression technique and doesn't make full use of the edge characteristics to predict the input vector. Classified side-match vector quantization (CSMVQ) is an effective image compression technique with low bit rate and relatively high reconstruction quality. It exploits a block classifier to decide which class the input vector belongs to using the variances of neighboring blocks' codewords. As an alternative, this paper proposes three algorithms using gradient values of neighboring blocks' codewords to predict the input block. The first one employs a basic gradient-based classifier that is similar to CSMVQ. To achieve lower bit rates, the second one exploits a refined two-level classifier structure. To reduce the encoding time further, the last one employs a more efficient classifier, in which adaptive class codebooks are defined within a gradient-ordered master codebook according to various prediction results. Experimental results prove the effectiveness of the proposed algorithms.
Gian Marco BO Daniele D. CAVIGLIA Maurizio VALLE
In this paper we present the analog architecture and the implementation of an on-chip learning Multi Layer Perceptron network. The learning algorithm is based on Back Propagation but it exhibits increased capabilities due to local learning rate management. A prototype chip (SLANP, Self-Learning Neural Processor) has been designed and fabricated in a CMOS 0.7 µm minimum channel length technology. We report the experimental results that confirm the functionality of the chip and the soundness of the approach. The SLANP performance compare favourably with those reported in the literature.
Min-Cheol HONG Hyung Tae CHA Hern-Soo HAHN
In this letter, we propose a spatially adaptive image restoration algorithm, using local statistics. The local variance, mean and maximum value are utilized to constrain the solution space. These parameters are computed at each iteration step using partially restored image. A parameter defined by the user determines the degree of local smoothness imposed on the solution. The resulting iterative algorithm exhibits increased convergence speed when compared with the non-adaptive algorithm. In addition, a smooth solution with a controlled degree of smoothness is obtained. Experimental results demonstrate the capability of the proposed algorithm.
Haiyun JIANG Shotaro NISHIMURA Takao HINAMOTO
In this paper, we present a method to analyze the steady-state performance of a complex coefficient adaptive IIR notch filter which is useful for the rejection of multiple narrow-band interferences from broad-band signals in quadrature phase shift keying (QPSK) spread-spectrum communication systems. The adaptive notch filter based on the simplified gradient algorithm is considered. Analytical expressions have been developed for the conditional mean and variance of notch filter output. The signal-to-noise ratio improvement factor is also obtained from which the validity of the use of the notch filter can be concluded. Finally, the results of computer simulations are shown which confirm the theoretical predictions.
Blagovest SHISHKOV Jun CHENG Takashi OHIRA
The electronically steerable passive array radiator (ESPAR) antenna performs analog aerial beamforming that has only a single-port output and none of the signals on its passive elements can be observed. This fact and one that is more important--the highly nonlinear dependence of the output of the antenna from adjustable reactances--makes the problem substantially new and not resolvable by means of conventional adaptive array beamforming techniques. A novel approach based on stochastic approximation theory is proposed for the adaptive beamforming of the ESPAR antenna as a nonlinear spatial filter by variable parameters, thus forming both beam and nulls. Our theoretic study, simulation results and performance analysis show that the ESPAR antenna can be controlled effectively, has strong potential for use in mobile terminals and seems to be very perspective.
A method of learning for multi-layer artificial neural networks is proposed. The learning model is designed to provide an effective means of escape from the Backpropagation local minima. The system is shown to escape from the Backpropagation local minima and be of much faster convergence than simulated annealing techniques by simulations on the exclusive-or problem and the Arabic numerals recognition problem.
Rong-Long WANG Zheng TANG Qi-Ping CAO
A near-optimum parallel algorithm for bipartite subgraph problem using gradient ascent learning algorithm of the Hopfield neural networks is presented. This parallel algorithm, uses the Hopfield neural network updating to get a near-maximum bipartite subgraph and then performs gradient ascent learning on the Hopfield network to help the network escape from the state of the near-maximum bipartite subgraph until the state of the maximum bipartite subgraph or better one is obtained. A large number of instances have been simulated to verify the proposed algorithm, with the simulation result showing that our algorithm finds the solution quality is superior to that of best existing parallel algorithm. We also test the proposed algorithm on maximum cut problem. The simulation results also show the effectiveness of this algorithm.
Yusuke KAWASAKI Naotaka NITTA Tsuyoshi SHIINA
Technique of Measuring 3-D velocity vector components is important for the correct diagnosis of the blood flow pattern and quantitative assessment of intratumor perfusion. However, present equipment based on ultrasonic Doppler can not provide us true 3-D velocity. To overcome the problem, we previously proposed a new method of 3-D velocity vector measurement. The method uses 2-D array probe and enable us to obtain three components of velocity vector with real time by integrating the Doppler phase shift on the each element with the relative small single aperture compared with conventional method. Basic performance of the method has been evaluated by computer simulation. In this paper, to evaluate the feasibility of the proposed method, experimental investigation using a simple ring array probe and a phantom were carried out. Three components of velocity vector for different velocity magnitude and flow direction were measured. Experimental results validated its ability of measuring 3-D velocity and its feasibility.
We study a class of nonlinear dynamical systems to develop efficient algorithms. As an efficient algorithm, interior point method based on Newton's method is well-known for solving convex programming problems which include linear, quadratic, semidefinite and lp-programming problems. On the other hand, the geodesic of information geometry is represented by a continuous Newton's method for minimizing a convex function called divergence. Thus, we discuss a relation between information geometry and convex programming in a related family of continuous Newton's method. In particular, we consider the α-projection problem from a given data onto an information geometric submanifold spanned with power-functions. In general, an information geometric structure can be induced from a standard convex programming problem. In contrast, the correspondence from information geometry to convex programming is slightly complicated. We first present there exists a same structure between the α-projection and semidefinite programming problems. The structure is based on the linearities or autoparallelisms in the function space and the space of matrices, respectively. However, the α-projection problem is not a form of convex programming. Thus, we reformulate it to a lp-programming and the related ones. For the reformulated problems, we derive self-concordant barrier functions according to the values of α. The existence of a polynomial time algorithm is theoretically confirmed for the problem. Furthermore, we present the coincidence with the gradient vectors for the divergence and a modified barrier function. These results connect a part of nonlinear and algorithm theories by the discreteness of variables.
Sungsoo AHN Seungwon CHOI Tapan K. SARKAR
This letter introduces an alternative adaptive beamforming with the total computational load of about O(3N) where N denotes the number of antenna elements. The proposed technique finds a weight vector that maximizes the received signal power at the array output by searching for the suboptimal phase of each weight in a multipath fading CDMA mobile communication environment.
Jun CHENG Yukihiro KAMIYA Takashi OHIRA
Conventional adaptive array antenna processing must access signals on all of the array antenna elements. However, because the low-cost electronically steerable passive array radiator (ESPAR) antenna only has a single-port output, all of the signals on the antenna elements cannot be observed. In this paper, a technique for adaptively controlling the loaded reactances on the passive radiators, thus forming both beam and nulls, is presented for the ESPAR antenna. The adaptive algorithm is based on the steepest gradient theory, where the reactances are sequentially perturbed to determine the gradient vector. Simulations show that the ESPAR antenna can be adaptive. The statistical performance of the output SIR of the ESPAR antenna is also given.
Zheng TANG Rong Long WANG Qi Ping CAO
A gradient ascent learning algorithm of the Hopfield neural networks for graph planarization is presented. This learning algorithm uses the Hopfield neural network to get a near-maximal planar subgraph, and increases the energy by modifying parameters in a gradient ascent direction to help the network escape from the state of the near-maximal planar subgraph to the state of the maximal planar subgraph or better one. The proposed algorithm is applied to several graphs up to 150 vertices and 1064 edges. The performance of our algorithm is compared with that of Takefuji/Lee's method. Simulation results show that the proposed algorithm is much better than Takefuji/Lee's method in terms of the solution quality for every tested graph.
Michiharu MAEDA Hiromi MIYAJIMA
This paper is concerned with fuzzy modeling in some reduction methods of inference rules with gradient descent. Reduction methods are presented, which have a reduction mechanism of the rule unit that is applicable in three parameters--the central value and the width of the membership function in the antecedent part, and the real number in the consequent part--which constitute the standard fuzzy system. In the present techniques, the necessary number of rules is set beforehand and the rules are sequentially deleted to the prespecified number. These methods indicate that techniques other than the reduction approach introduced previously exist. Experimental results are presented in order to show that the effectiveness differs between the proposed techniques according to the average inference error and the number of learning iterations.
Information geometry is applied to the manifold of neural networks called multilayer perceptrons. It is important to study a total family of networks as a geometrical manifold, because learning is represented by a trajectory in such a space. The manifold of perceptrons has a rich differential-geometrical structure represented by a Riemannian metric and singularities. An efficient learning method is proposed by using it. The parameter space of perceptrons includes a lot of algebraic singularities, which affect trajectories of learning. Such singularities are studied by using simple models. This poses an interesting problem of statistical inference and learning in hierarchical models including singularities.
Seungjin CHOI Shunichi AMARI Andrzej CICHOCKI
Spatio-temporal decorrelation is the task of eliminating correlations between associated signals in spatial domain as well as in time domain. In this paper, we present a simple but efficient adaptive algorithm for spatio-temporal decorrelation. For the task of spatio-temporal decorrelation, we consider a dynamic recurrent network and calculate the associated natural gradient for the minimization of an appropriate optimization function. The natural gradient based spatio-temporal decorrelation algorithm is applied to the task of blind deconvolution of linear single input multiple output (SIMO) system and its performance is compared to the spatio-temporal anti-Hebbian learning rule.
First order line seach optimization techniques gained essential practical importance over second order optimization techniques due to their computational simplicity and low memory requirements. The computational excess of second order methods becomes unbearable for large optimization tasks. The only applicable optimization techniques in such cases are variations of first order approaches. This article presents one such variation of first order line search optimization technique. The presented algorithm has substantially simplified a line search subproblem into a single step calculation of the appropriate value of step length. This remarkably simplifies the implementation and computational complexity of the line search subproblem and yet does not harm the stability of the method. The algorithm is theoretically proven convergent, with superlinear convergence rates, and exactly classified within the formerly proposed classification framework for first order optimization. Performance of the proposed algorithm is practically evaluated on five data sets and compared to the relevant standard first order optimization technique. The results indicate superior performance of the presented algorithm over the standard first order method.
This article addresses two issues. Firstly, the convergence property of conjugate gradient (CG) algorithm is investigated by a Chebyshev polynomial approximation. The analysis result shows that its convergence behaviour is affected by an acceleration term over the steepest descent (SD) algorithm. Secondly, a new CG algorithm is proposed in order to boost the tracking capability for time-varying parameters. The proposed algorithm based on re-initialising forgetting factor shows a fast tracking ability and a noise-immunity property when it encounters an unexpected parameter change. A fast tracking capability is verified through a computer simulation in a system identification problem.
Songyot SUREERATTANAN Huynh Ngoc PHIEN
A new algorithm is proposed for improving the convergence of backpropagtion networks. This algorithm is obtained by combining the conjugate gradient method and the Kalman filter algorithm. Simulation results show that the proposed algorithm can perform satisfactorily in all cases considered.