We introduce a novel algorithm that computes the $k$-sparse principal component of a positive semidefinite matrix $A$. Our algorithm is combinatorial and operates by examining a discrete set of special vectors lying in a low-dimensional eigen-subspace of $A$. We obtain provable approximation guarantees that depend on the spectral decay profile of the matrix: the faster the eigenvalue decay, the better the quality of our approximation. For example, if the eigenvalues of $A$ follow a power-law decay, we obtain a polynomial-time approximation algorithm for any desired accuracy. A key algorithmic component of our scheme is a combinatorial feature elimination step that is provably safe and in practice significantly reduces the running complexity of our algorithm. We implement our algorithm and test it on multiple artificial and real data sets. Due to the feature elimination step, it is possible to perform sparse PCA on data sets consisting of millions of entries in a few minutes. Our experimental evaluation shows that our scheme is nearly optimal while finding very sparse vectors. We compare to the prior state of the art and show that our scheme matches or outperforms previous algorithms in all tested data sets.

Many problems can be formulated as recovering a low-rank tensor. Although an increasingly common task, tensor recovery remains a challenging problem because of the delicacy associated with the decomposition of higher order tensors. To overcome these difficulties, existing approaches often proceed by unfolding tensors into matrices and then apply techniques for matrix completion. We show here that such matricization fails to exploit the tensor structure and may lead to suboptimal procedure. More specifically, we investigate a convex optimization approach to tensor completion by directly minimizing a tensor nuclear norm and prove that this leads to an improved sample size requirement. To establish our results, we develop a series of algebraic and probabilistic techniques such as characterization of subdifferetial for tensor nuclear norm and concentration inequalities for tensor martingales, which may be of independent interests and could be useful in other tensor related problems.

When I first encountered PAC-Bayesian concentration inequalities they seemed to me to be rather disconnected from good old-fashioned results like Hoeffding's and Bernstein's inequalities. But, at least for one flavour of the PAC-Bayesian bounds, there is actually a very close relation, and the main innovation is a continuous version of the union bound, along with some ingenious applications. Here's the gist of what's going on, presented from a machine learning perspective.

Social media systems rely on user feedback and rating mechanisms for personalization, ranking, and content filtering. However, when users evaluate content contributed by fellow users (e.g., by liking a post or voting on a comment), these evaluations create complex social feedback effects. This paper investigates how ratings on a piece of content affect its author's future behavior. By studying four large comment-based news communities, we find that negative feedback leads to significant behavioral changes that are detrimental to the community. Not only do authors of negatively-evaluated content contribute more, but also their future posts are of lower quality, and are perceived by the community as such. Moreover, these authors are more likely to subsequently evaluate their fellow users negatively, percolating these effects through the community. In contrast, positive feedback does not carry similar effects, and neither encourages rewarded authors to write more, nor improves the quality of their posts. Interestingly, the authors that receive no feedback are most likely to leave a community. Furthermore, a structural analysis of the voter network reveals that evaluations polarize the community the most when positive and negative votes are equally split.

We present a machine learning framework for modeling protein dynamics. Our approach uses L1-regularized, reversible hidden Markov models to understand large protein datasets generated via molecular dynamics simulations. Our model is motivated by three design principles: (1) the requirement of massive scalability; (2) the need to adhere to relevant physical law; and (3) the necessity of providing accessible interpretations, critical for both cellular biology and rational drug design. We present an EM algorithm for learning and introduce a model selection criteria based on the physical notion of convergence in relaxation timescales. We contrast our model with standard methods in biophysics and demonstrate improved robustness. We implement our algorithm on GPUs and apply the method to two large protein simulation datasets generated respectively on the NCSA Bluewaters supercomputer and the Folding@Home distributed computing network. Our analysis identifies the conformational dynamics of the ubiquitin protein critical to cellular signaling, and elucidates the stepwise activation mechanism of the c-Src kinase protein.

A new approach to maximum likelihood learning of discrete graphical models and RBM in particular is introduced. Our method, Perturb and Descend (PD) is inspired by two ideas (I) perturb and MAP method for sampling (II) learning by Contrastive Divergence minimization. In contrast to perturb and MAP, PD leverages training data to learn the models that do not allow efficient MAP estimation. During the learning, to produce a sample from the current model, we start from a training data and descend in the energy landscape of the "perturbed model", for a fixed number of steps, or until a local optima is reached. For RBM, this involves linear calculations and thresholding which can be very fast. Furthermore we show that the amount of perturbation is closely related to the temperature parameter and it can regularize the model by producing robust features resulting in sparse hidden layer activation.

We consider the question of learning in general topological vector spaces. By exploiting known (or parametrized) covariance structures, our Main Theorem demonstrates that any continuous linear map corresponds to a certain isomorphism of embedded Hilbert spaces. By inverting this isomorphism and extending continuously, we construct a version of the Ordinary Least Squares estimator in absolute generality. Our Gauss-Markov theorem demonstrates that OLS is a "best linear unbiased estimator", extending the classical result. We construct a stochastic version of the OLS estimator, which is a continuous disintegration exactly for the class of "uncorrelated implies independent" (UII) measures. As a consequence, Gaussian measures always exhibit continuous disintegrations through continuous linear maps, extending a theorem of the first author. Applying this framework to some problems in machine learning, we prove a useful representation theorem for covariance tensors, and show that OLS defines a good kriging predictor for vector-valued arrays on general index spaces. We also construct a support-vector machine classifier in this setting. We hope that our article shines light on some deeper connections between probability theory, statistics and machine learning, and may serve as a point of intersection for these three communities.

We propose a penalized likelihood method to jointly estimate multiple precision matrices for use in quadratic discriminant analysis and model based clustering. A ridge penalty and a ridge fusion penalty are used to introduce shrinkage and promote similarity between precision matrix estimates. Block-wise coordinate descent is used for optimization, and validation likelihood is used for tuning parameter selection. Our method is applied in quadratic discriminant analysis and semi-supervised model based clustering.

Rare data in a large-scale database are called outliers that reveal significant information in the real world. The subspace-based outlier detection is regarded as a feasible approach in very high dimensional space. However, the outliers found in subspaces are only part of the true outliers in high dimensional space, indeed. The outliers hidden in normal-clustered points are sometimes neglected in the projected dimensional subspace. In this paper, we propose a robust subspace method for detecting such inner outliers in a given dataset, which uses two dimensional-projections: detecting outliers in subspaces with local density ratio in the first projected dimensions; finding outliers by comparing neighbor's positions in the second projected dimensions. Each point's weight is calculated by summing up all related values got in the two steps projected dimensions, and then the points scoring the largest weight values are taken as outliers. By taking a series of experiments with the number of dimensions from 10 to 10000, the results show that our proposed method achieves high precision in the case of extremely high dimensional space, and works well in low dimensional space.

We consider the problem of minimizing the regret in stochastic multi-armed bandit, when the measure of goodness of an arm is not the mean return, but some general function of the mean and the variance. We characterize the conditions under which learning is possible and present examples for which no natural algorithm can achieve sublinear regret.