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.

Multi-objective artificial immune algorithm for fuzzy clustering based on multiple kernels

Abstract: This paper presents a multi-objective artificial immune algorithm for fUzzy clustering based on multiple kernels (MAFC) MAFC extends the classical Fuzzy C-Means (FCM) algorithm and overcomes its important limitations, such as limited adaptability, poor handling of non-linear relationships between data, and vulnerability to local optima convergence, which can lead to poor clustering quality To compensate these limitations, MAFC unifies multi-kernel learning and multi-objective optimization in a joint clustering framework, which preserves the geometric information of the dataset The multikernel method maps data from the feature space to kernel space by kernel functions This approach is effective, not only for spherical clusters, but can also discover the non-linear relationships between data, and adds robustness to the particular choice of kernel functions Additionally, the introduction of multi-objective optimization can optimize between-cluster separation and within-cluster compactness simultaneously via two different clustering validity criteria These properties help the proposed algorithm to avoid becoming stuck at local optima Furthermore, this paper utilizes an artificial immune algorithm to address the multi-objective clustering problem and acquire a Pareto optimal solution set The solution set is obtained through the process of antibody population initialization, clone proliferation, non-uniform mutation and uniformity maintaining strategy, which avoids the problems of degradation and prematurity which can occur with conventional genetic algorithms Finally, we choose the best solution, from the Pareto optimal solution set, using a semi-supervised method, to achieve the final clustering results We compare our method against three state-of-the-art methods from the literature by performing experiments with both UCI datasets and face datasets The results suggest that MAFC is significantly more efficient for clustering and has a wider scope of application

Cross-Corpus Speech Emotion Recognition Based on Multiple Kernel Learning of Joint Sample and Feature Matching

Abstract: Cross-corpus speech emotion recognition, which learns an accurate classifier for new test data using old and labeled training data, has shown promising value in speech emotion recognition research. Most previous works have explored two learning strategies independently for cross-corpus speech emotion recognition: feature matching and sample reweighting. In this paper, we show that both strategies are important and inevitable when the distribution difference is substantially large for training and test data. We therefore put forward a novel multiple kernel learning of joint sample and feature matching (JSFM-MKL) to model them in a unified optimization problem. Experimental results demonstrate that the proposed JSFM-MKL outperforms the competitive algorithms for cross-corpus speech emotion recognition.

Facial Expression Analysis via Transfer Learning

Abstract: Automated analysis of facial expressions has remained an interesting and challenging research topic in the field of computer vision and pattern recognition due to vast applications such as human-machine interface design, social robotics, and developmental psychology. This dissertation focuses on developing and applying transfer learning algorithms – multiple kernel learning (MKL) and multi-task learning (MTL) – to resolve the problems of facial feature fusion and the exploitation of multiple facial action units (AUs) relations in designing robust facial expression recognition systems. MKL algorithms are employed to fuse multiple facial features with different kernel functions and tackle the domain adaption problem at the kernel level within support vector machines (SVM). lp-norm is adopted to enforce both sparse and non-sparse kernel combination in our methods. We further develop and apply MTL algorithms for simultaneous detection of multiple related AUs by exploiting their inter-relationships. Three variants of task structure models are designed and investigated to obtain fine depiction of AU relations. lp-norm MTMKL and TDMTMKL (Task-Dependent MTMKL) are group-sensitive MTL methods that model the co-occurrence relations among AUs. On the other hand, our proposed hierarchical multi-task structural learning (HMTSL) includes a latent layer to learn a hierarchical structure to exploit all possible AU inter-relations for AU detection. Extensive ex-

Multiple Kernel Maximum Margin Criterion

Abstract: Maximum Margin Criterion (MMC) is an efficient and robust feature extraction method, which has been proposed recently. Like other kernel methods, when MMC is extended to Reproducing Kernel Hilbert Space via kernel trick, its performance heavily depends on the choice of kernel. In this paper, we address the problem of learning the optimal kernel over a convex set of prescribed kernels for Kernel MMC (KMMC). We will give an equivalent graph based formulation of MMC, based on which we present Multiple Kernel Maximum Margin Criterion (MKMMC). Then we will show that MKMMC can be solved via alternative optimization schema. Experiments on benchmark image recognition data sets show that the proposed method outperforms KMMC via cross validation, as well as some state of the art methods.