This paper surveys various implementation of a memory efficient second order (Broyden, Fletcher, Goldfard and Shanno) BFGS training algorithms which includes novel optimal memory (OM) BFGS neural network training algo...
详细信息
This paper surveys various implementation of a memory efficient second order (Broyden, Fletcher, Goldfard and Shanno) BFGS training algorithms which includes novel optimal memory (OM) BFGS neural network training algorithm, proposed by the present authors, which optimises performance in relation to available memory. Simulation results using a control benchmark problems show that OM BFGS, which is mathematically equivalent to full memory (FM) BFGS training when there are no constraints on memory, have performance gain compared to other memory efficient BFGS training algorithms.
This work investigates novel sequential learning methods applied on a decomposed form of training algorithms using radial basis function (RBF) network. The dynamic expansion of RBF network by adding neurons to the hid...
详细信息
This work investigates novel sequential learning methods applied on a decomposed form of training algorithms using radial basis function (RBF) network. The dynamic expansion of RBF network by adding neurons to the hidden layer during the course of training facilitates the weight update to be decomposed on neuron by neuron basis. The fast or minimal update approach which can be adopted with ease on a decomposed algorithms are also presented in This work.
This paper presents a novel sequential learning neural network implementation of action dependent adaptive critics. Sequential learning neural networks provide a systematic way of adding neurons in response to new dat...
详细信息
This paper presents a novel sequential learning neural network implementation of action dependent adaptive critics. Sequential learning neural networks provide a systematic way of adding neurons in response to new data features as well as removing neurons which cease to contribute to the overall performance of the network. The convergence rate of the sequential learning method is enhanced by applying a modified Recursive Prediction Error algorithm to adjust network parameters. The new methodology, which provides a fully autonomous controller, is benchmarked against the conventional MLP neurocontroller on a highly nonlinear inverted pendulum system and shown to achieve superior perfonnance.
Augmented reality is the merging of synthetic sensory information into a user's perception of a real environment. As one of the most important tasks in augmented scene modeling, terrain simplification research has...
详细信息
Augmented reality is the merging of synthetic sensory information into a user's perception of a real environment. As one of the most important tasks in augmented scene modeling, terrain simplification research has gained more and more attention. In this paper, we mainly focus on point selection problem in terrain simplification using triangulated irregular network. Based on the analysis and comparison of traditional importance measures for each input point, we put forward a new importance measure based on local entropy. The results demonstrate that the local entropy criterion has a better performance than any traditional methods. In addition, it can effectively conquer the 'short-sight' problem associated with the traditional methods.
Because of wide variation in gray levels and particle dimensions and the presence of many small gravel objects in the background, as well as corrupting the image by noise, it is difficult o segment gravel objects. In ...
详细信息
Because of wide variation in gray levels and particle dimensions and the presence of many small gravel objects in the background, as well as corrupting the image by noise, it is difficult o segment gravel objects. In this paper, we develop a partial entropy method and succeed to realize gravel objects segmentation. We give entropy principles and fur calculation methods. Moreover, we use minimum entropy error automaticly to select a threshold to segment image. We introduce the filter method using mathematical morphology. The segment experiments are performed by using different window dimensions for a group of gravel image and demonstrates that this method has high segmentation rate and low noise sensitivity.
This paper presents knowledge based approach to structure level adaptation of neural network. This algorithm determines a network structure based on prior knowledge and generates and/or annihilates hidden neurons of t...
详细信息
This paper presents knowledge based approach to structure level adaptation of neural network. This algorithm determines a network structure based on prior knowledge and generates and/or annihilates hidden neurons of the network to reach good structure during learning phase. Furthermore, we present a method of extraction of fuzzy rules from the regularities of the network, since the network structure is one of optimal network structures. To verify the effectiveness of the proposed method, we developed a model of the occurrence of hypertension and extracted fuzzy rules from the network.
control methods based on using the relative motion between the manipulator and the workpiece are described for controlling the force of a one-dimensional manipulator, in which it is assumed that there are no collision...
详细信息
control methods based on using the relative motion between the manipulator and the workpiece are described for controlling the force of a one-dimensional manipulator, in which it is assumed that there are no collisions between the manipulator and the workpiece and we use a computed force law which is similar to the computed torque law in the trajectory tracking problem of a manipulator. We consider two cases depending on whether the position and velocity of the workpiece (or end-effector) are available or not to calculate the computed force control. The effectiveness of the proposed control methods is illustrated by some computer simulations.
暂无评论