Statistical analysis of kernel-based least-squares density-ratio estimation

The ratio of two probability densities can be used for solving various machine learning tasks such as covariate shift adaptation (importance sampling), outlier detection (likelihood-ratio test), feature selection (mutual information), and conditional probability estimation. Several methods of directly estimating the density ratio have recently been developed, e.g., moment matching estimation, maximum-likelihood density-ratio estimation, and least-squares density-ratio fitting. In this paper, we propose a kernelized variant of the least-squares method for density-ratio estimation, which is called kernel unconstrained least-squares importance fitting (KuLSIF). We investigate its fundamental statistical properties including a non-parametric convergence rate, an analytic-form solution, and a leave-one-out cross-validation score. We further study its relation to other kernel-based density-ratio estimators. In experiments, we numerically compare various kernel-based density-ratio estimation methods, and show that KuLSIF compares favorably with other approaches.     

Semi-supervised learning combining co-training with active learning

Co-training is a good paradigm of semi-supervised, which requires the data set to be described by two views of features. There are a notable characteristic shared by many co-training algorithm: the selected unlabeled instances should be predicted with high confidence, since a high confidence score usually implies that the corresponding prediction is correct. Unfortunately, it is not always able to improve the classification performance with these high confidence unlabeled instances. In this paper, a new semi-supervised learning algorithm was proposed combining the benefits of both co-training and active learning. The algorithm applies co-training to select the most reliable instances according to the two criterions of high confidence and nearest neighbor for boosting the classifier, also exploit the most informative instances with human annotation for improve the classification performance. Experiments on several UCI data sets and natural language processing task, which demonstrate our method achieves more significant improvement for sacrificing the same amount of human effort.     

Semi-supervised learning with regularized Laplacian

We study a semi-supervised learning method based on the similarity graph and regularized Laplacian. We give convenient optimization formulation of the regularized Laplacian method and establish its various properties. In particular, we show that the kernel of the method can be interpreted in terms of discrete and continuous-time random walks and possesses several important properties of proximity measures. Both optimization and linear algebra methods can be used for efficient computation of the classification functions. We demonstrate on numerical examples that the regularized Laplacian method is robust with respect to the choice of the regularization parameter and outperforms the Laplacian-based heat kernel methods.     

Semi-supervised learning with an imperfect supervisor

Real-life applications may involve huge data sets with misclassified or partially classified training data. Semi-supervised learning and learning in the presence of label noise have recently emerged as new paradigms in the machine learning community to cope with this kind of problems. This paper describes a new discriminant algorithm for semi-supervised learning. This algorithm optimizes the classification maximum likelihood (CML) of a set of labeled–unlabeled data, using a discriminant extension of the Classification Expectation Maximization algorithm. We further propose to extend this algorithm by modeling imperfections in the estimated class labels for unlabeled data. The parameters of this label-error model are learned together with the semi-supervised classifier parameters. We demonstrate the effectiveness of the approach using extensive experiments on different datasets. Massih R. Amini is currently assistant professor in the University of Pierre and Marie Curie (Paris 6). He received an engineering degree in computer science from the Ecole Supérieure d'Informatique (Computer science engineering school) in Paris in 1995. He then accomplished his master thesis in science in artificial intelligence in 1997 and obtained his PhD in 2001 at University of Pierre and Marie Curie. His research interests include Statistical Learning and Text-Mining. Patrick Gallinari is currently professor in the University of Pierre and Marie Curie (Paris 6) and head of the Computer Science laboratory (LIP6). His main research activity has been in the field of statistical machine learning for the last 15 years. He has also contributed in developing machine learning techniques for different application domains like information retrieval and text mining, user modelling, man–machine interaction and pen interfaces.     

Semi-supervised learning by disagreement

In many real-world tasks, there are abundant unlabeled examples but the number of labeled training examples is limited, because labeling the examples requires human efforts and expertise. So, semi-supervised learning which tries to exploit unlabeled examples to improve learning performance has become a hot topic. Disagreement-based semi-supervised learning is an interesting paradigm, where multiple learners are trained for the task and the disagreements among the learners are exploited during the semi-supervised learning process. This survey article provides an introduction to research advances in this paradigm.     

Computational complexity of kernel-based density-ratio estimation: a condition number analysis

In this study, the computational properties of a kernel-based least-squares density-ratio estimator are investigated from the viewpoint of condition numbers. The condition number of the Hessian matrix of the loss function is closely related to the convergence rate of optimization and the numerical stability. We use smoothed analysis techniques and theoretically demonstrate that the kernel least-squares method has a smaller condition number than other M-estimators. This implies that the kernel least-squares method has desirable computational properties. In addition, an alternate formulation of the kernel least-squares estimator that possesses an even smaller condition number is presented. The validity of the theoretical analysis is verified through numerical experiments.     

Semi-supervised Unified Latent Factor learning with multi-view data

Explosive multimedia resources are generated on web, which can be typically considered as a kind of multi-view data in nature. In this paper, we present a Semi-supervised Unified Latent Factor learning approach (SULF) to learn a predictive unified latent representation by leveraging both complementary information among multiple views and the supervision from the partially label information. On one hand, SULF employs a collaborative Nonnegative Matrix Factorization formulation to discover a unified latent space shared across multiple views. On the other hand, SULF adopts a regularized regression model to minimize a prediction loss on partially labeled data with the latent representation. Consequently, the obtained parts-based representation can have more discriminating power. In addition, we also develop a mechanism to learn the weights of different views automatically. To solve the proposed optimization problem, we design an effective iterative algorithm. Extensive experiments are conducted for both classification and clustering tasks on three real-world datasets and the compared results demonstrate the superiority of our approach.     

Semi-supervised clustering with deep metric learning and graph embedding

World Wide Web - As a common technology in social network, clustering has attracted lots of research interest due to its high performance, and many clustering methods have been presented. The most...     

Semi-supervised dictionary learning with label propagation for image classification

Computational Visual Media - Sparse coding and supervised dictionary learning have rapidly developed in recent years, and achieved impressive performance in image classification. However, there is...     

Semi-supervised tensor learning for image classification

In this paper, we propose a new tensor-based representation algorithm for image classification. The algorithm is realized by learning the parameter tensor for image tensors. One novelty is that the parameter tensor is learned according to the Tucker tensor decomposition as the multiplication of a core tensor with a group of matrices for each order, which endows that the algorithm preserved the spatial information of image. We further extend the proposed tensor algorithm to a semi-supervised framework, in order to utilize both labeled and unlabeled images. The objective function can be solved by using the alternative optimization method, where at each iteration, we solve the typical ridge regression problem to obtain the closed form solution of the parameter along the corresponding order. Experimental results of gray and color image datasets show that our method outperforms several classification approaches. In particular, we find that our method can implement a high-quality classification performance when only few labeled training samples are provided.     

Semi-supervised learning using hidden feature augmentation

Semi-supervised learning methods are conventionally conducted by simultaneously utilizing abundant unlabeled samples and a few labeled samples given. However, the unlabeled samples are usually adopted with assumptions, e.g., cluster and manifold assumptions, which degrade the performance when the assumptions become invalid. The reliable hidden features embedded in both the labeled and the unlabeled samples can potentially be used to tackle this issue. In this regard, we investigate the feature augmentation technique to improve the robustness of semi-supervised learning in this paper. By introducing an orthonormal projection matrix, we first transform both the unlabeled and labeled samples into a shared hidden subspace to determine the connections between the samples. Then we utilize the hidden features, the raw features, and zero vectors determined to develop a novel feature augmentation strategy. Finally, a hidden feature transformation (HTF) model is proposed to compute the desired projection matrix by applying the maximum joint probability distribution principle in the augmented feature space. The effectiveness of the proposed method is evaluated in terms of the hinge and square loss functions respectively, based on two types of semi-supervised classification formulations developed using only the labeled samples with their original features and hidden features. The experimental results have demonstrated the effectiveness of the proposed feature augmentation technique for semi-supervised learning.     

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.     

Semi-supervised clustering with metric learning: An adaptive kernel method

Most existing representative works in semi-supervised clustering do not sufficiently solve the violation problem of pairwise constraints. On the other hand, traditional kernel methods for semi-supervised clustering not only face the problem of manually tuning the kernel parameters due to the fact that no sufficient supervision is provided, but also lack a measure that achieves better effectiveness of clustering. In this paper, we propose an adaptive Semi-supervised Clustering Kernel Method based on Metric learning (SCKMM) to mitigate the above problems. Specifically, we first construct an objective function from pairwise constraints to automatically estimate the parameter of the Gaussian kernel. Then, we use pairwise constraint-based K-means approach to solve the violation issue of constraints and to cluster the data. Furthermore, we introduce metric learning into nonlinear semi-supervised clustering to improve separability of the data for clustering. Finally, we perform clustering and metric learning simultaneously. Experimental results on a number of real-world data sets validate the effectiveness of the proposed method.     

Semi-supervised learning integrated with classifier combination for word sense disambiguation

Word sense disambiguation (WSD) is the problem of determining the right sense of a polysemous word in a certain context. This paper investigates the use of unlabeled data for WSD within a framework of semi-supervised learning, in which labeled data is iteratively extended from unlabeled data. Focusing on this approach, we first explicitly identify and analyze three problems inherently occurred piecemeal in the general bootstrapping algorithm; namely the imbalance of training data, the confidence of new labeled examples, and the final classifier generation; all of which will be considered integratedly within a common framework of bootstrapping. We then propose solutions for these problems with the help of classifier combination strategies. This results in several new variants of the general bootstrapping algorithm. Experiments conducted on the English lexical samples of Senseval-2 and Senseval-3 show that the proposed solutions are effective in comparison with previous studies, and significantly improve supervised WSD.     

半监督学习在不平衡样本集分类中的应用研究

在对不平衡样本集进行分类时容易产生少数类样误差大的问题,而目前半监督学习中的算法多数是针对未有明显此类特征的数据集。针对一种半监督协同分类算法在该问题上的有效性进行了研究。由于进一步增强了分类器差异性,该算法在理论上对不平衡样本集具有良好的分类性能。根据该算法建立分类模型,利用其对桥梁结构健康数据进行分类实验,与Tri-Training算法的结果比较表明,该算法对不平衡样本集具有良好的适用性,从而验证了上述算法的有效性。     

基于半监督对抗学习的图像语义分割

目的 将半监督对抗学习应用于图像语义分割,可以有效减少训练过程中人工生成标记的数量。作为生成器的分割网络的卷积算子只具有局部感受域,因此对于图像不同区域之间的远程依赖关系只能通过多个卷积层或增加卷积核的大小进行建模,但这种做法也同时失去了使用局部卷积结构获得的计算效率。此外,生成对抗网络(generative adversarial network, GAN)中的另一个挑战是判别器的性能控制。在高维空间中,由判别器进行的密度比估计通常是不准确且不稳定的。为此,本文提出面向图像语义分割的半监督对抗学习方法。方法 在生成对抗网络的分割网络中附加两层自注意模块,在空间维度上对语义依赖关系进行建模。自注意模块通过对所有位置的特征进行加权求和,有选择地在每个位置聚合特征。因而能够在像素级正确标记值数据的基础上有效处理输入图像中广泛分离的空间区域之间的关系。同时,为解决提出的半监督对抗学习方法的稳定性问题,在训练过程中将谱归一化应用到对抗网络的判别器中,这种加权归一化方法不仅可以稳定判别器网络的训练,并且不需要对唯一的超参数进行密集调整即可获得满意性能,且实现简单,计算量少,即使在缺乏互补的正则化...     

基于标签进行度量学习的图半监督学习算法

大多基于图的半监督学习方法，在样本间相似性度量时没有用到已有的和标签传播过程中得到的标签信息，同时，其度量方式相对固定，不能有效度量出分布结构复杂多样的数据样本间的相似性。针对上述问题，提出了基于标签进行度量学习的图半监督学习算法。首先，给定样本间相似性的度量方式，从而构建相似度矩阵。然后，基于相似度矩阵进行标签传播，筛选出k个低熵样本作为新确定的标签信息。最后，充分利用所有标签信息更新相似性度量方式，重复迭代优化直至学出所有标签信息。所提算法不仅利用标签信息改进了样本间相似性的度量方式，而且充分利用中间结果降低了半监督学习对标签数据的需求量。在6个真实数据集上的实验结果表明，该算法在超过95%的情况下相较三种传统的基于图的半监督学习算法取得了更高的分类准确率。     

基于标签进行度量学习的图半监督学习算法

大多基于图的半监督学习方法,在样本间相似性度量时没有用到已有的和标签传播过程中得到的标签信息,同时,其度量方式相对固定,不能有效度量出分布结构复杂多样的数据样本间的相似性。针对上述问题,提出了基于标签进行度量学习的图半监督学习算法。首先,给定样本间相似性的度量方式,从而构建相似度矩阵。然后,基于相似度矩阵进行标签传播,筛选出k个低熵样本作为新确定的标签信息。最后,充分利用所有标签信息更新相似性度量方式,重复迭代优化直至学出所有标签信息。所提算法不仅利用标签信息改进了样本间相似性的度量方式,而且充分利用中间结果降低了半监督学习对标签数据的需求量。在6个真实数据集上的实验结果表明,该算法在超过95%的情况下相较三种传统的基于图的半监督学习算法取得了更高的分类准确率。     

SELM: Semi-supervised ELM with application in sparse calibrated location estimation

Indoor location estimation based on Wi-Fi has attracted more and more attention from both research and industry fields. It brings two significant challenges. One is requiring a vast amount of labeled calibration data. The other is real-time training and testing for location estimation task. Traditional machine learning methods cannot get high performance in both aspects. This paper proposed a novel semi-supervised learning method SELM (semi-supervised extreme learning machine) and applied it to sparse calibrated location estimation. There are two advantages of the proposed SELM. First, it employs graph Laplacian regularization to import large number of unlabeled samples which can dramatically reduce labeled calibration samples. Second, it inherits the good property of ELM on extreme training and testing speed. Comparative experiments show that with same number of labeled samples, our method outperforms original ELM and back propagation (BP) network, especially in the case that the calibration data is very sparse.     

基于u-wordMixup的半监督深度学习模型

当标注样本匮乏时,半监督学习利用大量未标注样本解决标注瓶颈的问题,但由于未标注样本和标注样本来自不同领域,可能造成未标注样本存在质量问题,使得模型的泛化能力变差,导致分类精度下降.为此,基于wordMixup方法,提出针对未标注样本进行数据增强的u-wordMixup方法,结合一致性训练框架和Mean Teacher模型,提出一种基于u-wordMixup的半监督深度学习模型(semi-supervised deep learning model based on u-wordMixup,SD-uwM).该模型利用u-wordMixup方法对未标注样本进行数据增强,在有监督交叉熵和无监督一致性损失的约束下,能够提高未标注样本质量,减少过度拟合.在AGNews、THUCNews和20 Newsgroups数据集上的对比实验结果表明,所提出方法能够提高模型的泛化能力,同时有效提高时间性能.