Citation Context

...adially Gaussian kernel , and the Laplacian kernel , with the kernel bandwidth. Another example which deserves attention in signal processing is the th degree polynomial kernel defined as , with . In =-=[32]-=-, we have extended the framework of the so-called kernel-based methods to time-frequency analysis, showing that some specific reproducing kernels allow these algorithms to operate in the time-frequenc...

Citation Context

... choice of a (more or less arbitrary) distance and the evaluation of associated distributions. A different possibility would be to look at the statistical decision problem from a learning perspective =-=[3, 6, 13]-=-, and to consider surrogates as defining a learning set. Extraction of suitable descriptors from this learning set may therefore allows to make use of the powerful machinery of kernel methods, and esp...

Citation Context

...ossible. Here, we use in Section 4 specific bidimensional features, but one can think of directly using general TF representations emerging from the use of SVM machinery for TimeFrequency, such as in =-=[16]-=-. Another possibility is to generalize the present approach to other forms of stationarity. This requires to define new specific stationarizing tools and signal representations, in the spirit of [17, ...

Citation Context

...r breakthrough in the field of pattern recognition. This has led to new algorithms, such as Support Vector Machines (SVM), with improved performance and lower computational cost [4]. In a recent work =-=[5]-=-, we extended this framework to time-frequency analysis and proposed a new class of powerful tools for non-stationary signal analysis and classification. In that case, reproducing kernels simply corre...

Citation Context

...onceptual simplicity and their outstanding performance [4]. Initiated by Vapnik’s Support Vector Machines (SVM) [5], they offer now a wide class of supervised and unsupervised learning algorithms. In =-=[6, 7, 8]-=-, the authors have shown how This work was supported by ANR-07-BLAN-0191-01 STARAC. the most effective and innovative learning machines can be tuned to operate in the time-frequency domain. The presen...

Citation Context

...ceptual simplicity and their outstanding performance [30]. Initiated by Vapnik’s Support Vector Machines (SVM) [36], they offer now a wide class of supervised and unsupervised learning algorithms. In =-=[17, 18, 19]-=-, the authors have shown how the most effective and innovative learning machines can be tuned to operate in the time-frequency domain. The present paper follows this line of research by taking advanta...