Multiple kernel learning

About: Multiple kernel learning is a research topic. Over the lifetime, 1630 publications have been published within this topic receiving 56082 citations.

Fast Convergence Rate of Multiple Kernel Learning with Elastic-net Regularization

Abstract: We investigate the learning rate of multiple kernel leaning (MKL) with elastic-net regularization, which consists of an $\ell_1$-regularizer for inducing the sparsity and an $\ell_2$-regularizer for controlling the smoothness We focus on a sparse setting where the total number of kernels is large but the number of non-zero components of the ground truth is relatively small, and prove that elastic-net MKL achieves the minimax learning rate on the $\ell_2$-mixed-norm ball Our bound is sharper than the convergence rates ever shown, and has a property that the smoother the truth is, the faster the convergence rate is

Parametric Models for Mutual Kernel Matrix Completion

Abstract: Recent studies utilize multiple kernel learning to deal with incomplete-data problem. In this study, we introduce new methods that do not only complete multiple incomplete kernel matrices simultaneously, but also allow control of the flexibility of the model by parameterizing the model matrix. By imposing restrictions on the model covariance, overfitting of the data is avoided. A limitation of kernel matrix estimations done via optimization of an objective function is that the positive definiteness of the result is not guaranteed. In view of this limitation, our proposed methods employ the LogDet divergence, which ensures the positive definiteness of the resulting inferred kernel matrix. We empirically show that our proposed restricted covariance models, employed with LogDet divergence, yield significant improvements in the generalization performance of previous completion methods.

Spontaneous Speech Emotion Recognition via Multiple Kernel Learning

Abstract: Speech emotion recognition has become an active topic in pattern recognition. Specifically, support vector machine (SVM) is an effective classifier due to the application of the nonlinear mapping function, which can map the data into high or ever infinite dimensional feature space. However, a single kernel function might not sufficient to describe the different properties of spontaneous speech emotion data and thus it can not produce a satisfactory decision function. To address this issue, we apply multiple kernel learning (MKL) algorithm to recognize the spontaneous speech emotion. The experimental results are evaluated on the spontaneous speech emotion database such as FAU Aibo database. Compared to SVM, MKL can achieve better performance on spontaneous speech emotion recognition.

Kernel methods for gene regulatory network inference

Abstract: New technologies in molecular biology, in particular dna microarrays, have greatly increased the quantity of available data. in this context, methods from mathematics and computer science have been actively developed to extract information from large datasets. in particular, the problem of gene regulatory network inference has been tackled using many different mathematical and statistical models, from the most basic ones (correlation, boolean or linear models) to the most elaborate (regression trees, bayesian models with latent variables). despite their qualities when applied to similar problems, kernel methods have scarcely been used for gene network inference, because of their lack of interpretability. in this thesis, two approaches are developed to obtain interpretable kernel methods. firstly, from a theoretical point of view, some kernel methods are shown to consistently estimate a transition function and its partial derivatives from a learning dataset. these estimations of partial derivatives allow to better infer the gene regulatory network than previous methods on realistic gene regulatory networks. secondly, an interpretable kernel methods through multiple kernel learning is presented. this method, called lockni, provides state-of-the-art results on real and realistically simulated datasets.

Efficient similarity derived from kernel-based transition probability

Abstract: Semi-supervised learning effectively integrates labeled and unlabeled samples for classification, and most of the methods are founded on the pair-wise similarities between the samples In this paper, we propose methods to construct similarities from the probabilistic viewpoint, whilst the similarities have so far been formulated in a heuristic manner such as by k-NN We first propose the kernel-based formulation of transition probabilities via considering kernel least squares in the probabilistic framework The similarities are consequently derived from the kernel-based transition probabilities which are efficiently computed, and the similarities are inherently sparse without applying k-NN In the case of multiple types of kernel functions, the multiple transition probabilities are also obtained correspondingly From the probabilistic viewpoint, they can be integrated with prior probabilities, ie, linear weights, and we propose a computationally efficient method to optimize the weights in a discriminative manner, as in multiple kernel learning The novel similarity is thereby constructed by the composite transition probability and it benefits the semi-supervised learning methods as well In the various experiments on semi-supervised learning problems, the proposed methods demonstrate favorable performances, compared to the other methods, in terms of classification performances and computation time