摘要 :
The relationships between artificial neural networks and graph theory are considered in detail. The applications of artificial neural networks to many difficult problems of graph theory, especially NP-complete problems, and the ap...
展开
The relationships between artificial neural networks and graph theory are considered in detail. The applications of artificial neural networks to many difficult problems of graph theory, especially NP-complete problems, and the applications of graph theory to artificial neural networks are discussed. For example graph theory is used to study the pattern classification problem on the discrete type feedforward neural networks, and the stability analysis of feedback artificial neural networks etc.
收起
摘要 :
Recently, 2-band interpolating wavelet transform has attracted much attention. It has the following several features: (ⅰ)The wavelet series transform coefficients of a signal in the multiresolution subspace are exactly consistent...
展开
Recently, 2-band interpolating wavelet transform has attracted much attention. It has the following several features: (ⅰ)The wavelet series transform coefficients of a signal in the multiresolution subspace are exactly consistent with its discrete wavelet transform coefficints; (ⅱ)good approximation performance; (ⅲ)efficiency in computation.However orthogonal 2-band compactly supported interpolating wavelet transform is only the first order. In order to overcome this shortcoming, the orthogonal M-band compactly supported interpolating wavelet basis is established. First, the unitary interpolating scaling filters of the length L=MK are characterized. Second, a scheme is given to design highorder unitary interpolating scaling filters. Third, a parameterization of the unitary interpolating scaling filters of the length L=4M is made. Fourth, the orthogonal 2-order and 3-order three-band compactly supported interpolating scaling functions are constructed. Finally, the properties of the orthogonal M-band c
收起
摘要 :
Linear phase is not possible for real valued FIR QMF, while linear phase FIR biorthogonal wavelet filter banks make the mean squared error of the constructed signal exceed that of the quantization error. W Lawton’ s method for co...
展开
Linear phase is not possible for real valued FIR QMF, while linear phase FIR biorthogonal wavelet filter banks make the mean squared error of the constructed signal exceed that of the quantization error. W Lawton’ s method for complex valued wavelets construction is extended to generate the complex valued compactly supported wavelet packets that are symmetrical and unitary orthogonal; then well-defined wavelet packets are chosen by the analysis remarks on their time-frequency characteristics. Since the traditional wavelel packets transform coefficients do not exactly represent the strength of signal components, a modified adaptive wavelets transform, group-normalized wavelet packet transform (GNWPT), is presented and utilized for target extraction from formidable clutter or noises with the time-frequency masking technique. The extended definition of lp-norm entropy improves the performance cf GNWPT. Similar method can also be applied to image enhancement, clutter and noise suppression, optimal detection
收起
摘要 :
By generalizing the learning rate parameter to a learning rate matrix, this paper proposes agrading learning algorithm for blind source separation. The whole learning process is divided into threestages: initial stage, capturing s...
展开
By generalizing the learning rate parameter to a learning rate matrix, this paper proposes agrading learning algorithm for blind source separation. The whole learning process is divided into threestages: initial stage, capturing stage and tracking stage. In different stages, different learning rates areused for each output component, which is determined by its dependency on other output components. Itis shown that the grading learning algorithm is equivariant and can keep the separating matrix from be-coming singular. Simulations show that the proposed algorithm can achieve faster convergence, bettersteady-state performance and higher numerical robustness, as compared with the existing algorithmsusing fixed, time-descending and adaptive learning rates.
收起