Metric learning has become a critical tool in many machine learning tasks. This paper focuses on learning an optimal Mahalanobis distance matrix (parameterized by a positive semi-definite matrix W ) in the setting of supervised learning. Recently, particular research attention has been attracted by low-rank metric learning, which requires that matrix W is dominated by a few large singular values. In the era of high feature dimensions, low-rank metric learning effectively reduces the storage and computation overheads. However, existing low-rank metric learning algorithms usually adopt sophisticated regularization (such as LogDet divergence) for encouraging matrix low-rankness, which unfortunately incur iterative computations of matrix SVD. In this paper, we tackle low-rank metric learning by enforcing fixed-rank constraint on the matrix W. We harness the Riemannian manifold geometry of the collection of fixed-rank matrices and devise a novel second-order Riemannian retraction operator. The proposed operator is efficient and ensures that W always resides on the manifold. Comprehensive numerical experiments conducted on benchmarks clearly suggest that the proposed algorithm is substantially superior or on par with the state-of-the-art in terms of k -NN classification accuracy. Moreover, the proposed manifold retraction operator can be also naturally applied in generic rank-constrained machine learning algorithms.

In this study, we investigate expected CP decomposition — a special case of CP decomposition in which a tensor to be decomposed is given as the sum or average of tensor samples X ( t ) for t = 1,..., T . To determine this decomposition, we develope stochastic-gradient-descent-type algorithms with four appealing features: efficient memory use, ability to work in an online setting, robustness of parameter tuning, and simplicity. Our theoretical analysis show that the solutions do not diverge to infinity for any initial value or step size. Experimental results confirm that our algorithms significantly outperform all existing methods in terms of accuracy. We also show that they can successfully decompose a large tensor, containing billion-scale nonzero elements.

Computational agents often need to learn policies that involve many control variables, e.g., a robot needs to control several joints simultaneously. Learning a policy with a high number of parameters, however, usually requires a large number of training samples. We introduce a reinforcement learning method for sample-efficient policy search that exploits correlations between control variables. Such correlations are particularly frequent in motor skill learning tasks. The introduced method uses Variational Inference to estimate policy parameters, while at the same time uncovering a low-dimensional latent space of controls. Prior knowledge about the task and the structure of the learning agent can be provided by specifying groups of potentially correlated parameters. This information is then used to impose sparsity constraints on the mapping between the high-dimensional space of controls and a lower-dimensional latent space. In experiments with a simulated bi-manual manipulator, the new approach effectively identifies synergies between joints, performs efficient low-dimensional policy search, and outperforms state-of-the-art policy search methods.

The convolutional neural network (ConvNet or CNN) has proven to be very successful in many tasks such as those in computer vision. In this conceptual paper, we study the generative perspective of the discriminative CNN. In particular, we propose to learn the generative FRAME (Filters, Random field, And Maximum Entropy) model using the highly expressive filters pre-learned by the CNN at the convolutional layers. We show that the learning algorithm can generate realistic and rich object and texture patterns in natural scenes. We explain that each learned model corresponds to a new CNN unit at a layer above the layer of filters employed by the model. We further show that it is possible to learn a new layer of CNN units using a generative CNN model, which is a product of experts model, and the learning algorithm admits an EM interpretation with binary latent variables.

We consider an online learning problem (classification or prediction) involving disparate sources of sequentially arriving data, whereby a user over time learns the best set of data sources to use in constructing the classifier by exploiting their similarity. We first show that, when (1) the similarity information among data sources is known, and (2) data from different sources can be acquired without cost, then a judicious selection of data from different sources can effectively enlarge the training sample size compared to using a single data source, thereby improving the rate and performance of learning; this is achieved by bounding the classification error of the resulting classifier. We then relax assumption (1) and characterize the loss in learning performance when the similarity information must also be acquired through repeated sampling. We further relax both (1) and (2) and present a cost-efficient algorithm that identifies a best crowd from a potentially large set of data sources in terms of both classifier performance and data acquisition cost. This problem has various applications, including online prediction systems with time series data of various forms, such as financial markets, advertisement and network measurement.

Multiple kernel k-means (MKKM) clustering aims to optimally combine a group of pre-specified kernels to improve clustering performance. However, we observe that existing MKKM algorithms do not sufficiently consider the correlation among these kernels. This could result in selecting mutually redundant kernels and affect the diversity of information sources utilized for clustering, which finally hurts the clustering performance. To address this issue, this paper proposes an MKKM clustering with a novel, effective matrix-induced regularization to reduce such redundancy and enhance the diversity of the selected kernels. We theoretically justify this matrix-induced regularization by revealing its connection with the commonly used kernel alignment criterion. Furthermore, this justification shows that maximizing the kernel alignment for clustering can be viewed as a special case of our approach and indicates the extendability of the proposed matrix-induced regularization for designing better clustering algorithms. As experimentally demonstrated on five challenging MKL benchmark data sets, our algorithm significantly improves existing MKKM and consistently outperforms the state-of-the-art ones in the literature, verifying the effectiveness and advantages of incorporating the proposed matrix-induced regularization.

Feature selection has been widely recognized as one of the key problems in data mining and machine learning community, especially for high-dimensional data with redundant information, partial noises and outliers. Recently, unsupervised feature selection attracts substantial research attentions since data acquisition is rather cheap today but labeling work is still expensive and time consuming. This is specifically useful for effective feature selection of clustering tasks. Recent works using sparse projection with pre-learned pseudo labels achieve appealing results; however, they generate pseudo labels with all features so that noisy and ineffective features degrade the cluster structure and further harm the performance of feature selection; besides, these methods suffer from complex composition of multiple constraints and computational inefficiency, e.g., eigen-decomposition. Differently, in this work we introduce consensus clustering for pseudo labeling, which gets rid of expensive eigen-decomposition and provides better clustering accuracy with high robustness. In addition, complex constraints such as non-negative are removed due to the crisp indicators of consensus clustering. Specifically, we propose one efficient formulation for our unsupervised feature selection by using the utility function and provide theoretical analysis on optimization rules and model convergence. Extensive experiments on several popular data sets demonstrate that our methods are superior to the most recent state-of-the-art works in terms of NMI.

Autoregressive integrated moving average (ARIMA) is one of the most popular linear models for time series forecasting due to its nice statistical properties and great flexibility. However, its parameters are estimated in a batch manner and its noise terms are often assumed to be strictly bounded, which restricts its applications and makes it inefficient for handling large-scale real data. In this paper, we propose online learning algorithms for estimating ARIMA models under relaxed assumptions on the noise terms, which is suitable to a wider range of applications and enjoys high computational efficiency. The idea of our ARIMA method is to reformulate the ARIMA model into a task of full information online optimization (without random noise terms). As a consequence, we can online estimation of the parameters in an efficient and scalable way. Furthermore, we analyze regret bounds of the proposed algorithms, which guarantee that our online ARIMA model is provably as good as the best ARIMA model in hindsight. Finally, our encouraging experimental results further validate the effectiveness and robustness of our method.

Active learning seeks to train the best classifier at the lowest annotation cost by intelligently picking the best examples to label. Traditional algorithms assume there is a single annotator and disregard the possibility of requesting additional independent annotations for a previously labeled example. However, relabeling examples is important, because all annotators make mistakes — especially crowdsourced workers, who have become a common source of training data. This paper seeks to understand the difference in marginal value between decreasing the noise of the training set via relabeling and increasing the size and diversity of the (noisier) training set by labeling new examples. We use the term re-active learning to denote this generalization of active learning. We show how traditional active learning methods perform poorly at re-active learning, present new algorithms designed for this important problem, formally characterize their behavior, and empirically show that our methods effectively make this tradeoff.

In this paper, we focus on improving the performance of the Nyström based kernel SVM. Although the Nyström approximation has been studied extensively and its application to kernel classification has been exhibited in several studies, there still exists a potentially large gap between the performance of classifier learned with the Nyström approximation and that learned with the original kernel. In this work, we make novel contributions to bridge the gap without increasing the training costs too much by proposing a refined Nyström based kernel classifier. We adopt a two-step approach that in the first step we learn a sufficiently good dual solution and in the second step we use the obtained dual solution to construct a new set of bases for the Nyström approximation to re-train a refined classifier. Our approach towards learning a good dual solution is based on a sparse-regularized dual formulation with the Nyström approximation, which can be solved with the same time complexity as solving the standard formulation. We justify our approach by establishing a theoretical guarantee on the error of the learned dual solution in the first step with respect to the optimal dual solution under appropriate conditions. The experimental results demonstrate that (i) the obtained dual solution by our approach in the first step is closer to the optimal solution and yields improved prediction performance; and (ii) the second step using the obtained dual solution to re-train the model further improves the performance.