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.

Text detection in images based on Multiple Kernel Learning

Abstract: Detecting text accurately is an essential requirement for text recognition. In this paper, we propose a method to automatically detect text information in images. We firstly find the candidates of text regions based on the analysis of connected components and extract textural features in these candidate regions. We apply Multiple Kernel Learning to train a classifier with an optimal combination of kernels. The classifier can be used to distinguish text from icons which might be included in region candidates. Our method has been successfully implemented in detecting text from the interface images of mobile phones. According to the experimental results, our method outperforms several typical SVM based methods.

Tensor Decomposition in Multiple Kernel Learning

Abstract: ii

Novelty generating machine

Abstract: Novelty detection is one of primary tasks in data mining and machine learning. The task is to differentiate unseen outliers from normal patterns. Though novelty detection has been well-studied for many years and has found a wide range of applications, identifying outliers is still very challenging because of the absence or scarcity of outliers. We observe several characteristics of outliers and normal patterns. First, normal patterns are usually grouped together and form some clusters in high density regions of the data. Second, outliers are very different from the normal patterns, and in turn these outliers are far away from the normal patterns. Third, the number of outliers is very small compared with the size of the dataset. Based on these observations, we can envisage that the decision boundary between outliers and normal patterns usually lies in some low density regions of the data, which is referred to as cluster assumption. The resultant optimization problem is in form of a mixed integer programming. Then, we present a cutting plane algorithm together with multiple kernel learning techniques to solve its convex relaxation. Moreover, we make use of the scarcity of outliers to find a violating solution in cutting plane algorithm.

Prediction of phosphorylation sites using multiple kernel learning

Abstract: Phosphorylation is one of the most important post translational modifications which regulate the activity of proteins. The problem of predicting phosphorylation sites is the first step of understanding various biological processes that initiate the actual function of proteins in each signaling pathway. Although many prediction methods using single or multiple features extracted from protein sequences have been proposed, systematic data integration approach has not been applied in order to improve the accuracy of predicting general phosphorylation sites. In this paper, we propose an optimal way of integrating multiple features in the framework of multiple kernel learning. We optimally combine seven kernels extracted from sequence, physico-chemical properties, pairwise alignment, and structural information. Using the data set of Phospho.ELM, the accuracy evaluated by 5-fold cross-validation reaches 85% for serine, 85% for threonine, and 81% for tyrosine. Our computational experiments show significant improvement in the performance of prediction relative to a single feature, or to the combined feature with equal weights. Moreover, our systematic integration method significantly improves the prediction preformance compared with the previous well-known methods.