For many real-world tagging problems, training labels are usually obtained through social tagging and are notoriously incomplete. Consequently, handling data with incomplete labels has become a difficult challenge, which usually leads to a degenerated performance on label prediction. To improve the generalization performance, in this paper, we first propose the Improved Cross-View learning (referred as ICVL) model, which considers both global and local patterns of label relationship to enrich the original label set. Further, by extending the ICVL model with an outlier detection mechanism, we introduce the Improved Cross-View learning with Outlier Detection (referred as ICVL-OD) model to remove the abnormal tags resulting from label enrichment. Extensive evaluations on three benchmark datasets demonstrate that ICVL and ICVL-OD outstand with superior performances in comparison with the competing methods.

Domain generalization aims to apply knowledge gained from multiple labeled source domains to unseen target domains. The main difficulty comes from the dataset bias: training data and test data have different distributions, and the training set contains heterogeneous samples from different distributions. Let X denote the features, and Y be the class labels. Existing domain generalization methods address the dataset bias problem by learning a domain-invariant representation h(X) that has the same marginal distribution P(h(X)) across multiple source domains. The functional relationship encoded in P(Y|X) is usually assumed to be stable across domains such that P(Y|h(X)) is also invariant. However, it is unclear whether this assumption holds in practical problems. In this paper, we consider the general situation where both P(X) and P(Y|X) can change across all domains. We propose to learn a feature representation which has domain-invariant class conditional distributions P(h(X)|Y). With the conditional invariant representation, the invariance of the joint distribution P(h(X),Y) can be guaranteed if the class prior P(Y) does not change across training and test domains. Extensive experiments on both synthetic and real data demonstrate the effectiveness of the proposed method.

We present a novel method for frequentist statistical inference in M-estimation problems, based on stochastic gradient descent (SGD) with a fixed step size: we demonstrate that the average of such SGD sequences can be used for statistical inference, after proper scaling. An intuitive analysis using the Ornstein-Uhlenbeck process suggests that such averages are asymptotically normal. To show the merits of our scheme, we apply it to both synthetic and real data sets, and demonstrate that its accuracy is comparable to classical statistical methods, while requiring potentially far less computation.

We consider a new setting of online clustering of contextual cascading bandits, an online learning problem where the underlying cluster structure over users is unknown and needs to be learned from a random prefix feedback. More precisely, a learning agent recommends an ordered list of items to a user, who checks the list and stops at the first satisfactory item, if any. We propose an algorithm of CLUB-cascade for this setting and prove an n-step regret bound of order O(√n). Previous work corresponds to the degenerate case of only one cluster, and our general regret bound in this special case also significantly improves theirs. We conduct experiments on both synthetic and real data, and demonstrate the effectiveness of our algorithm and the advantage of incorporating online clustering method.

Many interesting problems in machine learning are being revisited with new deep learning tools. For graph-based semi-supervised learning, a recent important development is graph convolutional networks (GCNs), which nicely integrate local vertex features and graph topology in the convolutional layers. Although the GCN model compares favorably with other state-of-the-art methods, its mechanisms are not clear and it still requires considerable amount of labeled data for validation and model selection. In this paper, we develop deeper insights into the GCN model and address its fundamental limits. First, we show that the graph convolution of the GCN model is actually a special form of Laplacian smoothing, which is the key reason why GCNs work, but it also brings potential concerns of over-smoothing with many convolutional layers. Second, to overcome the limits of the GCN model with shallow architectures, we propose both co-training and self-training approaches to train GCNs. Our approaches significantly improve GCNs in learning with very few labels, and exempt them from requiring additional labels for validation. Extensive experiments on benchmarks have verified our theory and proposals.

Sparse and low rank coding has widely received much attention in machine learning, multimedia and computer vision. Unfortunately, expensive inference restricts the power of coding models in real-world applications, e.g., compressed sensing and image deblurring. In order to avoid the expensive inference, we propose a predictive coding machine (PCM) which aims to train a deep neural network (DNN) encoder to approximate the codes. By this means, a test sample can be fast approximated by the well-trained DNN. However, DNN leads PCM to be a non-convex and non-smooth optimization problem, which is extremely hard to solve. To address this challenge, we extend accelerated proximal gradient for PCM by steering gradient descent of DNN. To the best of our knowledge, we are the first to propose a gradient descent algorithm guided by accelerated proximal gradient for solving the PCM problem. Besides, a sufficient condition is provided to ensure the convergence to a critical point. Moreover, when the coding models are convex in PCM, the convergence rate O (1/( m 2 √ t )) can be held in which m is the iteration number of accelerated proximal gradient, and t is the epoch of training DNN. Numerical results verify the promising advantages of PCM in terms of effectiveness, efficiency and robustness.

Some recent works in classification show that the data obtained from various views with different sensors for an object contributes to achieving a remarkable performance. Actually, in many real-world applications, each view often contains multiple features, which means that this type of data has a hierarchical structure, while most of existing works do not take these features with multi-layer structure into consideration simultaneously. In this paper, a probabilistic hierarchical model is proposed to address this issue and applied for classification. In our model, a latent variable is first learned to fuse the multiple features obtained from a same view, sensor or modality. Particularly, mapping matrices corresponding to a certain view are estimated to project the latent variable from a shared space to the multiple observations. Since this method is designed for the supervised purpose, we assume that the latent variables associated with different views are influenced by their ground-truth label. In order to effectively solve the proposed method, the Expectation-Maximization (EM) algorithm is applied to estimate the parameters and latent variables. Experimental results on the extensive synthetic and two real-world datasets substantiate the effectiveness and superiority of our approach as compared with state-of-the-art.

A classic approach toward zero-shot learning (ZSL) is to map the input domain to a set of semantically meaningful attributes that could be used later on to classify unseen classes of data (e.g. visual data). In this paper, we propose to learn a visual feature dictionary that has semantically meaningful atoms. Such a dictionary is learned via joint dictionary learning for the visual domain and the attribute domain, while enforcing the same sparse coding for both dictionaries. Our novel attribute aware formulation provides an algorithmic solution to the domain shift/hubness problem in ZSL. Upon learning the joint dictionaries, images from unseen classes can be mapped into the attribute space by finding the attribute aware joint sparse representation using solely the visual data. We demonstrate that our approach provides superior or comparable performance to that of the state of the art on benchmark datasets.

Imitation learning refers to the problem where an agent learns a policy that mimics the demonstration provided by the expert, without any information on the cost function of the environment. Classical approaches to imitation learning usually rely on a restrictive class of cost functions that best explains the expert's demonstration, exemplified by linear functions of pre-defined features on states and actions. We show that the kernelization of a classical algorithm naturally reduces the imitation learning to a distribution learning problem, where the imitation policy tries to match the state-action visitation distribution of the expert. Closely related to our approach is the recent work on leveraging generative adversarial networks (GANs) for imitation learning, but our reduction to distribution learning is much simpler, robust to scarce expert demonstration, and sample efficient. We demonstrate the effectiveness of our approach on a wide range of high-dimensional control tasks.

The vanishing ideal is a set of polynomials that takes zero value on the given data points. Originally proposed in computer algebra, the vanishing ideal has been recently exploited for extracting the nonlinear structures of data in many applications. To avoid overfitting to noisy data, the polynomials are often designed to approximately rather than exactly equal zero on the designated data. Although such approximations empirically demonstrate high performance, the sound algebraic structure of the vanishing ideal is lost. The present paper proposes a vanishing ideal that is tolerant to noisy data and also pursued to have a better algebraic structure. As a new problem, we simultaneously find a set of polynomials and data points for which the polynomials approximately vanish on the input data points, and almost exactly vanish on the discovered data points. In experimental classification tests, our method discovered much fewer and lower-degree polynomials than an existing state-of-the-art method. Consequently, our method accelerated the runtime of the classification tasks without degrading the classification accuracy.