Yoshio Mogami

A hybrid algorithm for finding the global minimum of error function of neural networks and its applications

Abstract Back propagation has often been applied to adapt artificial neural networks for various pattern classification problems. However, an important limitation of this method is that it sometimes fails to find a global minimum of the total error function of the neural networks. In this article, a hybrid algorithm that combines the modified back-propagation method and the random optimization method is proposed to find the global minimum of the total error function of a neural network in a small number of steps. It is shown that this hybrid algorithm ensures convergence to a global minimum with probability 1 in a compact region of a weight vector space. Further, the results of several computer simulations dealing with the problems of forecasting air pollution density, forecasting stock prices, and determining the octane rating in gasoline blending are given.

A new learning algorithm for the hierarchical structure learning automata operating in the nonstationary S-model random environment

An extended algorithm of the relative reward strength algorithm is proposed. It is shown that the proposed algorithm ensures the convergence with probability I to the optimal path under the certain type of nonstationary environment. Several computer simulation results confirm the effectiveness of the proposed algorithm.

A relative reward-strength algorithm for the hierarchical structure learning automata operating in the general nonstationary multiteacher environment

A new learning algorithm for the hierarchical structure learning automata (HSLA) operating in the nonstationary multiteacher environment (NME) is proposed. The proposed algorithm is derived by extending the original relative reward-strength algorithm to be utilized in the HSLA operating in the general NME. It is shown that the proposed algorithm ensures convergence with probability 1 to the optimal path under a certain type of the NME. Several computer-simulation results, which have been carried out in order to compare the relative performance of the proposed algorithm in some NMEs against those of the two of the fastest algorithms today, confirm the effectiveness of the proposed algorithm

Continuous-time model identification by using adaptive observer

Abstract This paper proposes a continuous-time model identification from sampled I/O data by using an adaptive observer, The boundedness of the parameter estimate and the exponential convergence of the parameter estimate error to 0 under the PE assumption are guaranteed, In order to identify the plant from a finite number of the I/O data, an adaptive observer of a backward system is also proposed.

Continuous-time model identification by using adaptive observer -Estimation of the initial state

Importance of the continuous-time model identification is increasing as the computer becomes faster. The main difference between the continuous-time model identification and the discrete-time model identification lies not only in the information loss incurred by sampling but also in the fact that the transient term must be taken into account in the continuous-time identification. This paper proposes a continuous-time model estimation method from sampled I/O data with the ZOH input assumption. The initial state of the plant is estimated by using a backward system of the plant. An upper bound of the plant degree is assumed to be known. Numerical example shows that the proposed method performs well as the conventional method with an initial state estimation

DISCRETE-TIME STATE VARIABLE FILTERS FOR THE ESTIMATION OF A CONTINUOUS-TIME MODEL

Abstract This paper proposes a continuous-time model estimation method from sampled I/O data with the ZOH input assumption. State variable filters are modified in order to estimate the intersample signals by using a previously obtained parameter estimate. A sufficient condition for the parameter convergence and the bias of the estimated parameters are analyzed.

A Consideration on the Learning Behaviors of the HSLA Under the Nonstationary Multiteacher Environment and Their Application to Simulation and Gaming

Learning behaviors of the hierarchical structure stochastic automata operating in the nonstationary multiteacher environment are considered. A new learning algorithm which extends the idea of the relative reward strength algorithm is proposed. It is shown that the proposed algorithm ensures convergence to the optimal path under a certain type of the nonstationary multiteacher environment. Learning behaviors of the proposed algorithm are simulated by computer and the results indicate its effectiveness.

Utilization of stochastic automata for neural network learning

Backpropagation method has been applied to various pattern classification problems. However, one of the most important limitations of this method is that it often fails to find a global minimum of total error function of neural networks. In order to overcome this limitation, me have recently proposed a hybrid algorithm which combines the random optimization method with the modified backpropagation method. This hybrid algorithm has been successfully applied to several actual problems, such as air pollution density forecasting, stock price forecasting, etc. In this paper, the learning performance of stochastic automaton is utilized to accelerate the convergence of this hybrid algorithm. Several computer simulation results confirm our ideas.

Bias Compensated State Space Identification Method in Closed Loop Environment

Abstract A method of bias compensation in subspace identification method in closed loop environment is proposed. The noise is assumed to be a 0 mean colored noise and is assumed to be uncorrelated with the reference input. The covariance matrix of the noise is estimated directly from the residuals instead of estimating the noise model. The proposing method becomes an iterative algorithm and it is analyzed that the proposing method achieves 2nd order convergence.

A sufficient condition for the convergence of adaptive observer in continuous-time model identification

Abstract This paper proposes a new method to identify a continuous-time model from sampled I/O data by using an adaptive observer. The boundedness of the parameter estimate and the exponential convergence of the parameter estimation error to 0 under the assumption of Persistently exciting (PE) signals are guaranteed. In order to identify the plant from a finite number of I/O data, an adaptive observer for a backward system is also proposed. In noise-free cases, the magnitude of the parameter estimation error can be made arbitrarily small if the sampling period is smaller than a certain value.