Semi-supervised kernel density estimation for video annotation

Insufficiency of labeled training data is a major obstacle for automatic video annotation. Semi-supervised learning is an effective approach to this problem by leveraging a large amount of unlabeled data. However, existing semi-supervised learning algorithms have not demonstrated promising results in large-scale video annotation due to several difficulties, such as large variation of video content and intractable computational cost. In this paper, we propose a novel semi-supervised learning algorithm named semi-supervised kernel density estimation (SSKDE) which is developed based on kernel density estimation (KDE) approach. While only labeled data are utilized in classical KDE, in SSKDE both labeled and unlabeled data are leveraged to estimate class conditional probability densities based on an extended form of KDE. It is a non-parametric method, and it thus naturally avoids the model assumption problem that exists in many parametric semi-supervised methods. Meanwhile, it can be implemented with an efficient iterative solution process. So, this method is appropriate for video annotation. Furthermore, motivated by existing adaptive KDE approach, we propose an improved algorithm named semi-supervised adaptive kernel density estimation (SSAKDE). It employs local adaptive kernels rather than a fixed kernel, such that broader kernels can be applied in the regions with low density. In this way, more accurate density estimates can be obtained. Extensive experiments have demonstrated the effectiveness of the proposed methods.     

Wavelet kernel learning

This paper addresses the problem of optimal feature extraction from a wavelet representation. Our work aims at building features by selecting wavelet coefficients resulting from signal or image decomposition on an adapted wavelet basis. For this purpose, we jointly learn in a kernelized large-margin context the wavelet shape as well as the appropriate scale and translation of the wavelets, hence the name “wavelet kernel learning”. This problem is posed as a multiple kernel learning problem, where the number of kernels can be very large. For solving such a problem, we introduce a novel multiple kernel learning algorithm based on active constraints methods. We furthermore propose some variants of this algorithm that can produce approximate solutions more efficiently. Empirical analysis show that our active constraint MKL algorithm achieves state-of-the art efficiency. When used for wavelet kernel learning, our experimental results show that the approaches we propose are competitive with respect to the state-of-the-art on brain–computer interface and Brodatz texture datasets.     

基于多核学习的医学文献蛋白质关系抽取

从生物医学文献中抽取蛋白质交互作用关系对蛋白质知识网络的建立、新药的研制等均具有重要的意义。为此,提出一种基于多核学习的方法,用于从文献中自动抽取蛋白质关系信息。该方法融合基于特征的核、树核以及图核,并扩展最短路径依存树以及依存路径以利用更多的上下文关系信息。在AImed语料上的实验得到63.9%的F值和87.83%的AUC值,表明该方法具有较好的性能。     

A novel approach to wavelet selection and tree kernel construction for diagnosis of rolling element bearing fault

A novel methodology for early diagnosis of rolling element bearing fault is employed based on continuous wavelet transform (CWT) and support vector machine (SVM). CWT is especially suited for analyzing non-stationary signals in time–frequency domain where time information is retained as well as frequency content. To better approximate non-stationary vibration signals from rolling element bearing, a wavelet choice criterion is established to select an appropriate mother wavelet for feature extraction. The Shannon wavelet is picked out of several considered wavelets. The classification tree kernels (CTK) are constructed to address nonlinear classification of the characteristic samples derived from the wavelet coefficients. By using Fuzzy pruning strategy, a large variety of classification trees are generated. The trees with diverse structures can effectively explore intrinsic information among samples. Then, the tree kernel matrices can be acquired through ensemble statistical learning, which eventually reveal the similarity of samples objectively and stably. Under such architecture of kernel methods, a classification tree kernel based support vector machine (CTKSVM) is proposed to identify bearing fault. The performance of the methodology involving CWT and CTKSVM (CWT–CTKSVM) is evaluated by cross validation and independent test. The results show that the CWT–CTKSVM totally is superior to other SVM methods with common kernels. Therefore, it is a prospective technique for detection and identification of rolling element bearing fault.     

A sparse kernel algorithm for online time series data prediction

Kernel based methods have been widely applied for signal analysis and processing. In this paper, we propose a sparse kernel based algorithm for online time series prediction. In classical kernel methods, the kernel function number is very large which makes them of a high computational cost and only applicable for off-line or batch learning. In online learning settings, the learning system is updated when each training sample is obtained and it requires a higher computational speed. To make the kernel methods suitable for online learning, we propose a sparsification method based on the Hessian matrix of the system loss function to continuously examine the significance of the new training sample in order to select a sparse dictionary (support vector set). The Hessian matrix is equivalent to the correlation matrix of sample inputs in the kernel weight updating using the recursive least square (RLS) algorithm. This makes the algorithm able to be easily implemented with an affordable computational cost for real-time applications. Experimental results show the ability of the proposed algorithm for both real-world and artificial time series data forecasting and prediction.     

Accelerated max-margin multiple kernel learning

Kernel machines such as Support Vector Machines (SVM) have exhibited successful performance in pattern classification problems mainly due to their exploitation of potentially nonlinear affinity structures of data through the kernel functions. Hence, selecting an appropriate kernel function, equivalently learning the kernel parameters accurately, has a crucial impact on the classification performance of the kernel machines. In this paper we consider the problem of learning a kernel matrix in a binary classification setup, where the hypothesis kernel family is represented as a convex hull of fixed basis kernels. While many existing approaches involve computationally intensive quadratic or semi-definite optimization, we propose novel kernel learning algorithms based on large margin estimation of Parzen window classifiers. The optimization is cast as instances of linear programming. This significantly reduces the complexity of the kernel learning compared to existing methods, while our large margin based formulation provides tight upper bounds on the generalization error. We empirically demonstrate that the new kernel learning methods maintain or improve the accuracy of the existing classification algorithms while significantly reducing the learning time on many real datasets in both supervised and semi-supervised settings.     

Multilayer Graph Node Kernels: Stacking While Maintaining Convexity

Nowadays, developing effective techniques able to deal with data coming from structured domains is becoming crucial. In this context kernel methods are the state-of-the-art tool widely adopted in real-world applications that involve learning on structured data. Contrarily, when one has to deal with unstructured domains, deep learning methods represent a competitive, or even better, choice. In this paper we propose a new family of kernels for graphs which exploits an abstract representation of the information inspired by the multilayer perceptron architecture. Our proposal exploits the advantages of the two worlds. From one side we exploit the potentiality of the state-of-the-art graph node kernels. From the other side we develop a multilayer architecture through a series of stacked kernel pre-image estimators, trained in an unsupervised fashion via convex optimization. The hidden layers of the proposed framework are trained in a forward manner and this allows us to avoid the greedy layerwise training of classical deep learning. Results on real world graph datasets confirm the quality of the proposal.     

Deep multilayer multiple kernel learning

Multiple kernel learning (MKL) approach has been proposed for kernel methods and has shown high performance for solving some real-world applications. It consists on learning the optimal kernel from one layer of multiple predefined kernels. Unfortunately, this approach is not rich enough to solve relatively complex problems. With the emergence and the success of the deep learning concept, multilayer of multiple kernel learning (MLMKL) methods were inspired by the idea of deep architecture. They are introduced in order to improve the conventional MKL methods. Such architectures tend to learn deep kernel machines by exploring the combinations of multiple kernels in a multilayer structure. However, existing MLMKL methods often have trouble with the optimization of the network for two or more layers. Additionally, they do not always outperform the simplest method of combining multiple kernels (i.e., MKL). In order to improve the effectiveness of MKL approaches, we introduce, in this paper, a novel backpropagation MLMKL framework. Specifically, we propose to optimize the network over an adaptive backpropagation algorithm. We use the gradient ascent method instead of dual objective function, or the estimation of the leave-one-out error. We test our proposed method through a large set of experiments on a variety of benchmark data sets. We have successfully optimized the system over many layers. Empirical results over an extensive set of experiments show that our algorithm achieves high performance compared to the traditional MKL approach and existing MLMKL methods.     

Regularizing multiple kernel learning using response surface methodology

In recent years, several methods have been proposed to combine multiple kernels using a weighted linear sum of kernels. These different kernels may be using information coming from multiple sources or may correspond to using different notions of similarity on the same source. We note that such methods, in addition to the usual ones of the canonical support vector machine formulation, introduce new regularization parameters that affect the solution quality and, in this work, we propose to optimize them using response surface methodology on cross-validation data. On several bioinformatics and digit recognition benchmark data sets, we compare multiple kernel learning and our proposed regularized variant in terms of accuracy, support vector count, and the number of kernels selected. We see that our proposed variant achieves statistically similar or higher accuracy results by using fewer kernel functions and/or support vectors through suitable regularization; it also allows better knowledge extraction because unnecessary kernels are pruned and the favored kernels reflect the properties of the problem at hand.     

Multiple-kernel-learning-based extreme learning machine for classification design

The extreme learning machine (ELM) is a new method for using single hidden layer feed-forward networks with a much simpler training method. While conventional kernel-based classifiers are based on a single kernel, in reality, it is often desirable to base classifiers on combinations of multiple kernels. In this paper, we propose the issue of multiple-kernel learning (MKL) for ELM by formulating it as a semi-infinite linear programming. We further extend this idea by integrating with techniques of MKL. The kernel function in this ELM formulation no longer needs to be fixed, but can be automatically learned as a combination of multiple kernels. Two formulations of multiple-kernel classifiers are proposed. The first one is based on a convex combination of the given base kernels, while the second one uses a convex combination of the so-called equivalent kernels. Empirically, the second formulation is particularly competitive. Experiments on a large number of both toy and real-world data sets (including high-magnification sampling rate image data set) show that the resultant classifier is fast and accurate and can also be easily trained by simply changing linear program.     

Efficient Sparse Kernel Feature Extraction Based on Partial Least Squares

The presence of irrelevant features in training data is a significant obstacle for many machine learning tasks. One approach to this problem is to extract appropriate features and, often, one selects a feature extraction method based on the inference algorithm. Here, we formalize a general framework for feature extraction, based on Partial Least Squares, in which one can select a user-defined criterion to compute projection directions. The framework draws together a number of existing results and provides additional insights into several popular feature extraction methods. Two new sparse kernel feature extraction methods are derived under the framework, called Sparse Maximal Alignment (SMA) and Sparse Maximal Covariance (SMC), respectively. Key advantages of these approaches include simple implementation and a training time which scales linearly in the number of examples. Furthermore, one can project a new test example using only k kernel evaluations, where k is the output dimensionality. Computational results on several real-world data sets show that SMA and SMC extract features which are as predictive as those found using other popular feature extraction methods. Additionally, on large text retrieval and face detection data sets, they produce features which match the performance of the original ones in conjunction with a Support Vector Machine.     

多尺度核方法的自适应序列学习及应用

多尺度核方法是当前核机器学习领域的一个热点。通常多尺度核的学习在多核处理时存在诸如多核平均组合、迭代学习时间长、经验选择合成系数等弊端。文中基于核目标度量规则,提出一种多尺度核方法的自适应序列学习算法,实现多核加权系数的自动快速求取。实验表明,该方法在回归精度、分类正确率方面比单核支持向量机方法结果更优,函数拟合与分类稳定性更强,证明该算法具有普遍适用性。     

Inexact graph matching based on kernels for object retrieval in image databases

In the framework of online object retrieval with learning, we address the problem of graph matching using kernel functions. An image is represented by a graph of regions where the edges represent the spatial relationships. Kernels on graphs are built from kernel on walks in the graph. This paper firstly proposes new kernels on graphs and on walks, which are very efficient for graphs of regions. Secondly we propose fast solutions for exact or approximate computation of these kernels. Thirdly we show results for the retrieval of images containing a specific object with the help of very few examples and counter-examples in the framework of an active retrieval scheme.     

Building species trees from larger parts of phylogenomic databases

Gene trees are leaf-labeled trees inferred from molecular sequences. Because of gene duplication events arising in genomes, some species host several copies of the same gene, hence individual gene trees usually have several leaves labeled with identical species names. Dealing with such multi-labeled gene trees (MUL trees) is a substantial problem in phylogenomics, e.g. current supertree methods do not handle MUL trees, which restricts studies aimed at building the Tree of Life to a very small core of mono-copy genes. We propose to tackle this problem by mainly transforming a collection of MUL trees into a collection of trees, each containing single copies of labels. To achieve that aim, we provide several fast algorithmic building stones and describe how they fit in a general framework to build a species tree. First, we propose to separately preprocess each MUL tree in order to remove its redundant parts with respect to speciation events. For this purpose, we present a tree isomorphism algorithm for MUL trees to reduce redundant parts of these trees. Second, we show how the speciation signal contained in a MUL tree can be represented by a linear set of triplets. When this set is topologically coherent (compatible), we show that it can be used to produce a single-copy gene tree to replace the MUL tree while preserving the information it contains on speciation events. As an alternative approach, we propose to extract from each MUL tree a maximum size subtree that is free of duplication events. The algorithms are finally applied in a supertree analysis of hogenom, a database of homologous genes from fully sequenced genomes.     

弹性多核学习

多核学习 (MKL) 的提出是为了解决多个核矩阵的融合问题, 多核学习求解关于多个核矩阵的最优的线性组合并同时解出对应于这个组合矩阵的支持向量机(SVM)问题. 现有的多核学习的框架倾向于寻找稀疏的组合系数, 但是当有信息的核的比例较高的时候, 对稀疏性的倾向会使得只有少量的核被选中而损失相当的分类信息. 在本文中, 我们提出了弹性多核学习的框架来实现自适应的多核学习. 弹性多核学习的框架利用了一个混合正则化函数来均衡稀疏性和非稀疏性, 多核学习和支持向量机问题都可以视作弹性多核学习的特殊情形. 基于针对多核学习的梯度下降法, 我们导出了针对弹性多核学习的梯度下降法. 仿真数据的结果显示了弹性多核学习方法相对多核学习和支持向量机的优势; 我们还进一步将弹性多核学习应用于基因集合分析问题并取得了有意义的结果; 最后, 我们比较研究了弹性多核学习与另一种利用了非稀疏思想的多核学习.     

XML tree structure compression using RePair

XML tree structures can conveniently be represented using ordered unranked trees. Due to the repetitiveness of XML markup these trees can be compressed effectively using dictionary-based methods, such as minimal directed acyclic graphs (DAGs) or straight-line context-free (SLCF) tree grammars. While minimal SLCF tree grammars are in general smaller than minimal DAGs, they cannot be computed in polynomial time unless P=NP$\mathsf{P}=\mathsf{NP}$. Here, we present a new linear time algorithm for computing small SLCF tree grammars, called TreeRePair, and show that it greatly outperforms the best known previous algorithm BPLEX. TreeRePair is a generalization to trees of Larsson and Moffat's RePair string compression algorithm. SLCF tree grammars can be used as efficient memory representations of trees. Using TreeRePair, we are able to produce the smallest queryable memory representation of ordered trees that we are aware of. Our investigations over a large corpus of commonly used XML documents show that tree traversals over TreeRePair grammars are 14 times slower than over pointer structures and 5 times slower than over succinct trees, while memory consumption is only 1/43 and 1/6, respectively. With respect to file compression we are able to show that a Huffman-based coding of TreeRePair grammars gives compression ratios comparable to the best known XML file compressors.     

Incremental online learning in high dimensions

Locally weighted projection regression (LWPR) is a new algorithm for incremental nonlinear function approximation in high-dimensional spaces with redundant and irrelevant input dimensions. At its core, it employs nonparametric regression with locally linear models. In order to stay computationally efficient and numerically robust, each local model performs the regression analysis with a small number of univariate regressions in selected directions in input space in the spirit of partial least squares regression. We discuss when and how local learning techniques can successfully work in high-dimensional spaces and review the various techniques for local dimensionality reduction before finally deriving the LWPR algorithm. The properties of LWPR are that it (1) learns rapidly with second-order learning methods based on incremental training, (2) uses statistically sound stochastic leave-one-out cross validation for learning without the need to memorize training data, (3) adjusts its weighting kernels based on only local information in order to minimize the danger of negative interference of incremental learning, (4) has a computational complexity that is linear in the number of inputs, and (5) can deal with a large number of-possibly redundant-inputs, as shown in various empirical evaluations with up to 90 dimensional data sets. For a probabilistic interpretation, predictive variance and confidence intervals are derived. To our knowledge, LWPR is the first truly incremental spatially localized learning method that can successfully and efficiently operate in very high-dimensional spaces.