In the psychological noise evaluation, the fuzziness caused by the human subjective judgment for the acoustical stimulus is inevitably obscure. From the above viewpoints, in this paper, the relationship between the physical sound insulation method and the psychological noise evaluation method for fluctuating random noise is considered, by using the fuzzy set theory. That is, A set of eight simplified patterns of the membership functions are first established by using the actually observed data for psychological impression. Next, a new systematical method for estimating and/or predicting the psychological impression by using the above simplified membership functions is proposed, in the case when the two physical characteristics of noise stimulus (the sound pressure level probability distribution and the power spectral density) being changed by the erection of the sound insulation barrier.
In this paper, the relation between the correction and the forgetting of weights in the learning algorithm is examined and selective learning algorithms with forgetting are newly proposed. First, the relation between the direction of the correction and the direction where the weight is forgotten is examined and a learning algorithm with forgetting based on the learning error is proposed. Concretely, the weight is selectively forgotten by introducing the index to evaluate the learning ability. Next, a learning algorithm with forgetting based on the activity of the hidden unit is proposed in order to improve the structuring ability. Finally, the comparison examination of such learning algorithms with forgetting is done from the viewpoint of the learning ability, the structuring ability, and generalization ability by using the logical function learning problem, the classification problem of irises, and the time series prediction problem and the effectiveness of the proposed learning algorithm with forgetting is shown.
This paper describes a method which gives new pattern descriptions of observed images for computer vision with robustness to perturbation in the observed images. Although experimental results support that biological pattern vision should extract information about the observed images from signals with a variety of spatial frequencies over its broad receptive field, it is difficult to implement such an extraction due to the uncertainty relationship between the signal and its Fourier Transform; therefore, the biological pattern vision is considered to possess a function for suppression of this uncertainty relationship. We regard this function as fusion and consider that this fusion brings the robustness to the biological pattern vision; then, we propose a method of generating pattern descriptions of the observed images via this fusion. We first describe this fusion as a mathematical constraint. We second introduce a stochastic model satisfying this constraint. We consequently arrive at a Linear Programming; therefore, the pattern descriptions of the observed images can effectively be generated by the simplex method. We have made experiments on pattern matching with the proposed pattern descriptions. Their results have demonstrated that the proposed pattern descriptions are robust to the perturbations in contrast to a conventional method.
Accurate prediction of travel time is useful for car drivers, and thus it is critical to develop a prediction system to provide the information. AVI (Automatic Vehicle Identification) system has been developed as a direct and precise measurement tool, however, because of the constrains of the installation points, a few dozen systems have been installed so far. On the other hand, we have more than eighty thousand vehicle detectors which collect traffic volumes at each point every five minutes. We note that vehicle detectors are not quite adequate for the prediction of travel time since the traffic volume alone does not lead to the travel time. This paper proposes a new method of predicting travel time in real time using efficiently the data collected from the detectors. The route is divided into several sections. Then, the travel time of each section is estimated by the approximated velocity where the approximation is done via the time-varying coefficients autoregressive model. In order to evaluate this approach, several field experiments were carried out. The results show that predicted travel time closely approximates the measured value.
In this paper we examine the noise-reducing property of continuous wavelet transform, and propose a concept of optimal wavelets in the sense of complete noise reduction for separable noises in the time-scale domain. We first discuss about the wavelet transform of stochastic processes, focusing mainly on certain properties of it, such as the stationarity preserving property and Parseval-like identity. We then show that the mean square power of separable noises in the time-scale domain is possible to be zero by inverse wavelet transform if the basic wavelet φ (t) is chosen properly. A numerical example is given for illustration.
In this paper, we propose a new autoassociative memory model which is derived from Cross-Coupled Hopfield Nets (CCHN). The CCHN is a modular neural network in which plural Hopfield networks are mutually connected via feedforward neural networks. The CCHN's architecture is determined by the following structural parameters : the number of modules, the numbers of units in the modules, the contribution of the module information processings and the interactions to the whole network information processing, and the module connectivity. If these parameters are changed, the network dynamics are also changed; therefore, it may be possible to implement a great number of autoassociative memories with different nature. Through some computer simulations, we will discuss a diversity of association properties in the proposed model.