scispace - formally typeset
Search or ask a question
Topic

Kernel adaptive filter

About: Kernel adaptive filter is a research topic. Over the lifetime, 8771 publications have been published within this topic receiving 142711 citations.


Papers
More filters
Journal ArticleDOI
01 May 2000
TL;DR: In this article, an adaptive inverse control algorithm is proposed for shock testing an arbitrary specimen using an electrodynamic actuator, which is used to ascertain whether the specimen can survive and continue to function under severe shock conditions.
Abstract: An adaptive inverse control algorithm is proposed for shock testing an arbitrary specimen using an electrodynamic actuator. The purpose is to ascertain whether the specimen can survive and continue to function under severe shock conditions. The main difficulty in shock control is that the specimen dynamics vary significantly and a control algorithm is required that adapts to the characteristics of a new specimen. The control algorithm used is the adaptive inverse control method which approximates an inverse model of the loaded shaker with a finite impulse response adaptive filter, such that the reference input is reproduced at the shaker output. The standard filtered-x least mean square control structure used in the adaptive inverse control algorithm is modified to a block-processing structure, with the frequency-domain adaptive filter as the adaptation algorithm. Practical results show that the filtered-x frequency-domain adaptive filter control algorithm allows convergence of the shaker output to the assigned reference shock pulse.

36 citations

Posted Content
TL;DR: The kernel mixture network is introduced, a new method for nonparametric estimation of conditional probability densities using neural networks that can be used to filter complex nonlinear and non-Gaussian signals defined on manifolds.
Abstract: This paper introduces the kernel mixture network, a new method for nonparametric estimation of conditional probability densities using neural networks. We model arbitrarily complex conditional densities as linear combinations of a family of kernel functions centered at a subset of training points. The weights are determined by the outer layer of a deep neural network, trained by minimizing the negative log likelihood. This generalizes the popular quantized softmax approach, which can be seen as a kernel mixture network with square and non-overlapping kernels. We test the performance of our method on two important applications, namely Bayesian filtering and generative modeling. In the Bayesian filtering example, we show that the method can be used to filter complex nonlinear and non-Gaussian signals defined on manifolds. The resulting kernel mixture network filter outperforms both the quantized softmax filter and the extended Kalman filter in terms of model likelihood. Finally, our experiments on generative models show that, given the same architecture, the kernel mixture network leads to higher test set likelihood, less overfitting and more diversified and realistic generated samples than the quantized softmax approach.

36 citations

Patent
30 May 1989
TL;DR: In this paper, a system and method for generating coefficients for use in a digital filter using an iterative adaptive process employing a least mean square process is described. But, the method is not suitable for the case of noisy data.
Abstract: There is disclosed a system and method for generating coefficients for use in a digital filter. The coefficients are generated utilizing an iterative adaptive process employing a least mean square process wherein the filter coefficients are updated by an amount during each iteration dependent upon the stochastic average of the gradient generated during prior iterations. The response of a filter standard to an applied input signal is combined with a response of the adaptive filter coefficients to generate, during each iteration, an error signal. If the error signal is less than a predetermined standard, the iterative process is stopped, and the last used filter coefficients are utilized as the final filter coefficients of the digital filter.

36 citations

Journal ArticleDOI
TL;DR: A novel adaptive median-based lifting filter for image de-noising which has been corrupted by homogeneous salt and pepper noise is proposed and it is found that this method outperforms many other algorithms and it can remove salt and Pepper noise with a noise level as high as 90%.
Abstract: In this paper, we propose a novel adaptive median-based lifting filter for image de-noising which has been corrupted by homogeneous salt and pepper noise. The median-based lifting filter removes the noise of the input image by calculating the median of the neighboring significant pixels. The algorithm for image noise removal uses the lifting scheme of the second-generation wavelets in conjunction with the proposed adaptive median-based lifting filter. The experimental results demonstrate the efficiency of the proposed method. The proposed algorithm is compared with all the basic filters, and it is found that our method outperforms many other algorithms and it can remove salt and pepper noise with a noise level as high as 90%. The algorithm works exceedingly well for all levels of noise, as illustrated in terms of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) measures.

36 citations

Journal ArticleDOI
TL;DR: A sensor fusion algorithm that integrates gyroscope and vision measurements using an adaptive complementary Kalman filter is proposed to estimate the attitude of a hybrid head tracker system and has better performance than the conventional algorithms in high-dynamic conditions and vision measurement fault case.
Abstract: In this paper, a sensor fusion algorithm that integrates gyroscope and vision measurements using an adaptive complementary Kalman filter is proposed to estimate the attitude of a hybrid head tracker system. In order to make the filter more tolerant to vision measurement fault and more robust to system dynamics, an adaptive fading filter is implemented to the sensor fusion filter, and fuzzy logic is applied to adjust the fading factor, which adapts a Kalman gain of the sensor fusion filter. For recognizing the dynamic condition of the system and vision measurement fault, the normalized square error of attitude and the norm of gyroscope output with designed membership functions are used. The performance of the proposed algorithm is evaluated by simulations. It is confirmed that the proposed algorithm has better performance than the conventional algorithms in high-dynamic conditions and vision measurement fault case.

36 citations


Network Information
Related Topics (5)
Feature extraction
111.8K papers, 2.1M citations
85% related
Control theory
299.6K papers, 3.1M citations
83% related
Optimization problem
96.4K papers, 2.1M citations
79% related
Convolutional neural network
74.7K papers, 2M citations
79% related
Image processing
229.9K papers, 3.5M citations
79% related
Performance
Metrics
No. of papers in the topic in previous years
YearPapers
202322
202251
202113
202020
201931
201844