The gradient method for training Elman networks with a finite training sample set is considered. Monotonicity of the error function in the iteration is shown. Weak and strong convergence results are proved, indicating...The gradient method for training Elman networks with a finite training sample set is considered. Monotonicity of the error function in the iteration is shown. Weak and strong convergence results are proved, indicating that the gradient of the error function goes to zero and the weight sequence goes to a fixed point, respectively. A numerical example is given to support the theoretical findings.展开更多
This paper deals with Hermite learning which aims at obtaining the target function from the samples of function values and the gradient values. Error analysis is conducted for these algorithms by means of approaches f...This paper deals with Hermite learning which aims at obtaining the target function from the samples of function values and the gradient values. Error analysis is conducted for these algorithms by means of approaches from convex analysis in the frame- work of multi-task vector learning and the improved learning rates are derived.展开更多
The capacity of mobile communication system is improved by using Voice Activity Detection (VAD) technology. In this letter, a novel VAD algorithm, SVAD algorithm based on Fuzzy Neural Network Knowledge Discovery (FNNK...The capacity of mobile communication system is improved by using Voice Activity Detection (VAD) technology. In this letter, a novel VAD algorithm, SVAD algorithm based on Fuzzy Neural Network Knowledge Discovery (FNNKD) method is proposed. The performance of SVAD algorithm is discussed and compared with traditional algorithm recommended by ITU G.729B in different situations. The simulation results show that the SVAD algorithm performs better.展开更多
In the training of feedforward neural networks, it is usually suggested that the initial weights should be small in magnitude in order to prevent premature saturation. The aim of this paper is to point out the other s...In the training of feedforward neural networks, it is usually suggested that the initial weights should be small in magnitude in order to prevent premature saturation. The aim of this paper is to point out the other side of the story： In some cases, the gradient of the error functions is zero not only for infinitely large weights but also for zero weights. Slow convergence in the beginning of the training procedure is often the result of sufficiently small initial weights. Therefore, we suggest that, in these cases, the initial values of the weights should be neither too large, nor too small. For instance, a typical range of choices of the initial weights might be something like （-0.4,-0.1） U （0.1,0.4）, rather than （-0.1,0.1） as suggested by the usual strategy. Our theory that medium size weights should be used has also been extended to a few commonly used transfer functions and error functions. Numerical experiments are carried out to support our theoretical findings.展开更多
基金the National Natural Science Foundation of China (No.10471017)
文摘The gradient method for training Elman networks with a finite training sample set is considered. Monotonicity of the error function in the iteration is shown. Weak and strong convergence results are proved, indicating that the gradient of the error function goes to zero and the weight sequence goes to a fixed point, respectively. A numerical example is given to support the theoretical findings.
基金supported by the National Natural Science Foundation of China(No.11471292)
文摘This paper deals with Hermite learning which aims at obtaining the target function from the samples of function values and the gradient values. Error analysis is conducted for these algorithms by means of approaches from convex analysis in the frame- work of multi-task vector learning and the improved learning rates are derived.
文摘The capacity of mobile communication system is improved by using Voice Activity Detection (VAD) technology. In this letter, a novel VAD algorithm, SVAD algorithm based on Fuzzy Neural Network Knowledge Discovery (FNNKD) method is proposed. The performance of SVAD algorithm is discussed and compared with traditional algorithm recommended by ITU G.729B in different situations. The simulation results show that the SVAD algorithm performs better.
基金Project supported by the National Natural Science Foundation of China (No. 11171367)the Fundamental Research Funds for the Central Universities,China
文摘In the training of feedforward neural networks, it is usually suggested that the initial weights should be small in magnitude in order to prevent premature saturation. The aim of this paper is to point out the other side of the story： In some cases, the gradient of the error functions is zero not only for infinitely large weights but also for zero weights. Slow convergence in the beginning of the training procedure is often the result of sufficiently small initial weights. Therefore, we suggest that, in these cases, the initial values of the weights should be neither too large, nor too small. For instance, a typical range of choices of the initial weights might be something like （-0.4,-0.1） U （0.1,0.4）, rather than （-0.1,0.1） as suggested by the usual strategy. Our theory that medium size weights should be used has also been extended to a few commonly used transfer functions and error functions. Numerical experiments are carried out to support our theoretical findings.
