China 
Africa 
Explore chapters and articles related to this topic
Anomaly Detection in a Fleet of Systems
Published in Ashok N. Srivastava, Jiawei Han, Machine Learning and Knowledge Discovery for Engineering Systems Health Management, 2016
A common practice is to use a convex combination (i.e., ∑λβλ=1) of various kernels which may be constructed on very different feature sets such as color, shape, texture, etc. Therefore, a major advantage of the multiple-kernel learning approach is its ability to incorporate more knowledge in the decision process while analyzing complex heterogeneous systems that involve various data sources and data structures.
A Feature-Level Fusion Scheme Based on Eigen Theory for Multimodal Biometrics
Published in IETE Technical Review, 2022
Wen-Shiung Chen, Ren-He Jeng, Yen-Feng Chen
Feature combination is carried out through either serial rule or parallel rule [9]. It is intractable to reduce the dimensionality of such high dimensional data. To solve this problem, two types of main methods, such as feature extraction [11–15] and feature selection [16,17], have been developed. Feature extraction method is capable of identifying the useful features by removing redundant, irrelevant or noisy data with some functions, while keeping the abundant information, such as principal component analysis (PCA) [11] and linear discriminant analysis (LDA) [12]. Moreover, multiple kernel learning (MKL) method [19] aims at constructing a kernel model in which the kernel is a linear combination of fixed base kernels. Metric learning (ML) [20] is another method for learning a distance function over objects. Hashing-based approximated nearest neighbour search methods [21], such as locality-sensitive hashing (LSH) [22,23], have been proposed to construct hash functions based on random or principal projections. Whereas feature selection method is derived from feature extraction method. Feature selection methods [16,17] involve obtaining an appropriate subset to represent the whole set, based on a rigorous mathematical derivation [18,24]. This kind of method may be either filter-based approach [25] or wrapper-based approach [26].
Impulse response function identification of linear mechanical systems based on Kautz basis expansion with multiple poles
Published in International Journal of Systems Science, 2018
Changming Cheng, Zhike Peng, Xingjian Dong, Wenming Zhang, Guang Meng
In addition, the idea of Kautz basis expansion with multiple poles is similar to multiple kernel learning and ensemble learning in the machine learning community. Multiple kernel learning (Bach, Lanckriet, & Jordan, 2004; Gönen & Alpaydın, 2011; Sonnenburg, Rätsch, Schäfer, & Schölkopf, 2006) searches for an optimal linear or non-linear combination of predefined base kernels that maximises a generalised performance measure. Multiple kernel learning provides the needed flexibility and also reflects the fact that practical learning problems often involve multiple sources. The multiple kernels often can yield better model estimates than single kernels (especially for systems with complex dynamics), since different kernels may correspond to different features or modes. Ensemble learning (Zhou, 2012; Zhou, Wu, & Tang, 2002; Zhou & Yu, 2005) tries to construct a set of learners and combine them to obtain better learning ability. The learning ability of an ensemble is often much stronger than that of single learners. Ensemble methods are attractive mainly because they are able to boost weak learners which are even just slightly better than random guess to strong learners which can make very accurate predictions (Zhou, 2016; Zhou, Jiang, & Chen, 2003). The Kautz basis expansion with single pole corresponds to the single kernel in multiple kernel learning or one base learner in ensemble learning, and the Kautz basis expansion with multiple poles corresponds to multiple kernels in multiple kernel learning or a number of base learners in ensemble learning, so the learning ability of Kautz basis expansion with multiple poles is often better than Kautz basis expansion with single pole, especially for mechanical systems under excitation with a wide range of frequency.