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.

Active multiple kernel learning for interactive 3D object retrieval systems

Abstract: An effective relevance feedback solution plays a key role in interactive intelligent 3D object retrieval systems In this work, we investigate the relevance feedback problem for interactive intelligent 3D object retrieval, with the focus on studying effective machine learning algorithms for improving the user's interaction in the retrieval task One of the key challenges is to learn appropriate kernel similarity measure between 3D objects through the relevance feedback interaction with users We address this challenge by presenting a novel framework of Active multiple kernel learning (AMKL), which exploits multiple kernel learning techniques for relevance feedback in interactive 3D object retrieval The proposed framework aims to efficiently identify an optimal combination of multiple kernels by asking the users to label the most informative 3D images We evaluate the proposed techniques on a dataset of over 10,000 3D models collected from the World Wide Web Our experimental results show that the proposed AMKL technique is significantly more effective for 3D object retrieval than the regular relevance feedback techniques widely used in interactive content-based image retrieval, and thus is promising for enhancing user's interaction in such interactive intelligent retrieval systems

A Novel Method for Objective Selection of Information Sources Using Multi-Kernel SVM and Local Scaling.

Abstract: Advancement on computer and sensing technologies has generated exponential growth in the data available for the development of systems that support decision-making in fields such as health, entertainment, manufacturing, among others. This fact has made that the fusion of data from multiple and heterogeneous sources became one of the most promising research fields in machine learning. However, in real-world applications, to reduce the number of sources while maintaining optimal system performance is an important task due to the availability of data and implementation costs related to processing, implementation, and development times. In this work, a novel method for the objective selection of relevant information sources in a multimodality system is proposed. This approach takes advantage of the ability of multiple kernel learning (MKL) and the support vector machines (SVM) classifier to perform an optimal fusion of data by assigning weights according to their discriminative value in the classification task; when a kernel is designed for representing each data source, these weights can be used as a measure of their relevance. Moreover, three algorithms for tuning the Gaussian kernel bandwidth in the classifier prediction stage are introduced to reduce the computational cost of searching for an optimal solution; these algorithms are an adaptation of a common technique in unsupervised learning named local scaling. Two real application tasks were used to evaluate the proposed method: the selection of electrodes for a classification task in Brain-Computer Interface (BCI) systems and the selection of relevant Magnetic Resonance Imaging (MRI) sequences for detection of breast cancer. The obtained results show that the proposed method allows the selection of a small number of information sources.

Combining feature selection and representation for speech emotion recognition

Abstract: In this paper, we propose a feature selection and representation combination method to generate discriminative features for speech emotion recognition. In feature selection stage, a Multiple Kernel Learning (MKL) based strategy is used to obtain the optimal feature subset. Specifically, features selected at least n times among 10-fold cross validation are collected to build a new feature subset named n-subset, then the n-subset resulting in the highest classification accuracy is viewed as the optimal one. In feature representation stage, the optimal feature subset is mapped to a hidden representation using a denoising autoencoder (DAE). The model parameters are learned by minimizing the squared error between the original and the reconstructed input. The hidden representation is then used as the final feature set in the MKL model for emotion recognition. Our experimental results show significant performance improvement compared to using the original features in both of the inner-corpus and cross-corpus scenarios.

Selecting Scales by Multiple Kernel Learning for Shape Diffusion Analysis

Abstract: Brain morphological abnormalities can typically be detected by advanced geometrical shape analysis techniques. Recently, shape diffusion methods have proved to be very effective in providing useful descriptions for brain classification purposes. In particular, they allow the analysis of such shapes at multiple scales, but the selection of the correct range of scales remains an open issue heavily affecting the performance of methods, and it needs to be estimated adaptively for different classes of shapes. In this paper, we focus on the diffusion scale selection in order to define a robust shape descriptor for brain classification. To this end, geometric features are extracted for each scale and the best feature combination is selected by employing \it multiple kernel learning (MKL). In the presented experiments, we compare the shape of Thalamic regions in order to discriminate between normal subjects and schizophrenic patients. We demonstrate that MKL allows to obtain classifiers which are more accurate with respect to other competing algorithms for schizophrenia detection. Moreover, using the weights computed by the MKL algorithm, we can select at which scale the features are more effective for schizophrenia classification.