Continuous-valued word embeddings learned by neural language models have recently been shown to capture semantic and syntactic information about words very well, setting performance records on several word similarity tasks. The best results are obtained by learning high-dimensional embeddings from very large quantities of data, which makes scalability of the training method a critical factor. We propose a simple and scalable new approach to learning word embeddings based on training log-bilinear models with noise-contrastive estimation. Our approach is simpler, faster, and produces better results than the current state-of-theart method. We achieve results comparable to the best ones reported, which were obtained on a cluster, using four times less data and more than an order of magnitude less computing time. We also investigate several model types and find that the embeddings learned by the simpler models perform at least as well as those learned by the more complex ones.

Overview This paper proposes an algorithm for learning general structured predictors (e.g., non-linear). This is done by replacing the structured hinge loss with its smooth dual LP relaxation and observing that optimizing over classifiers reduces to a logistic regression task. Therefore, the learning problem can be extended to cases where this optimization over the class of predictors can be solved efficiently. Specifically, the paper shows how this enables learning predictors like decision trees and multi-layer perceptrons in addition to the common linear classifiers. Pros * The observation made by the authors about reduction of the learning objective to a logistic regression problem seems novel and interesting.

A successful approach to structured learning is to write the learning objective as a joint function of linear parameters and inference messages, and iterate between updates to each. This paper observes that if the inference problem is "smoothed" through the addition of entropy terms, for fixed messages, the learning objective reduces to a traditional (non-structured) logistic regression problem with respect to parameters. In these logistic regression problems, each training example has a bias term determined by the current set of messages. Based on this insight, the structured energy function can be extended from linear factors to any function class where an "oracle" exists to minimize a logistic loss.

This paper proposes determinantal point processes as a method to model inhibitory interactions in spike train data. The authors present a maximum likelihood approach based on stochastic gradient descent. This is an interesting idea that could potentially be a powerful non-factorial spike train model. However, the presentation here seems a bit unfocused, and it's not obvious to what extent DPPs will actually improve model accuracy. Since gain and periodic terms can easily be built into GLMs, it wasn't obvious to me that framing the problem as a DPP was worth the trouble.

Principal component analysis (PCA), a well-established technique for data analysis and processing, provides a convenient form of dimensionality reduction that is effective for cleaning small Gaussian noises presented in the data. However, the applicability of standard principal component analysis in real scenarios is limited by its sensitivity to large errors. In this paper, we tackle the challenge problem of recovering data corrupted with errors of high magnitude by developing a novel robust transfer principal component analysis method. Our method is based on the assumption that useful information for the recovery of a corrupted data matrix can be gained from an uncorrupted related data matrix. Specifically, we formulate the data recovery problem as a joint robust principal component analysis problem on the two data matrices, with common principal components shared across matrices and individual principal components specific to each data matrix. The formulated optimization problem is a minimization problem over a convex objective function but with non-convex rank constraints. We develop an efficient proximal projected gradient descent algorithm to solve the proposed optimization problem with convergence guarantees. Our empirical results over image denoising tasks show the proposed method can effectively recover images with random large errors, and significantly outperform both standard PCA and robust PCA with rank constraints.

This paper gives minimax lower bound for the number of bits, B, needed to be communicated between m estimators, each of which has acessess to n iid samples, such that the estimation error is on par with the centralized case (with mn samples). This is a very interesting problem of theoretic and practical significance. In view of the fact that the minimax rate without communication constraints which does not admit a general solution, I do not expect a general solution for the minimal B. As the authors shows the conclusion seems problem-specific and can be quite pessimistic, meaning a large amount of communication overhead is necessary. While the proof of Thm 2 is quite trivial (a vanilla application of Fano's inequality), the proof of Thm 3 is quite interesting which relies on a type of strong data processing inequality for mutual information under some bounded density condition, which is reminiscient of differential privacy and Thm 1 in J. C. Duchi, M. I. Jordan and M. J. Wainwright (2013). Whether one need even bigger B in order to reach the rate at sample size mn - I wonder if there is any problem in multi-user information theory that is similar in spirit, where a joint decoder has access to rate-constrained coded messages and want to recover the original source with some fidelity.

We establish lower bounds on minimax risks for distributed statistical estimation under a communication budget. Such lower bounds reveal the minimum amount of communication required by any procedure to achieve the centralized minimax-optimal rates for statistical estimation. We study two classes of protocols: one in which machines send messages independently, and a second allowing for interactive communication. We establish lower bounds for several problems, including various types of location models, as well as for parameter estimation in regression models.

In modeling multivariate time series, it is important to allow time-varying smoothness in the mean and covariance process. In particular, there may be certain time intervals exhibiting rapid changes and others in which changes are slow. If such locally adaptive smoothness is not accounted for, one can obtain misleading inferences and predictions, with over-smoothing across erratic time intervals and under-smoothing across times exhibiting slow variation. This can lead to miscalibration of predictive intervals, which can be substantially too narrow or wide depending on the time. We propose a continuous multivariate stochastic process for time series having locally varying smoothness in both the mean and covariance matrix. This process is constructed utilizing latent dictionary functions in time, which are given nested Gaussian process priors and linearly related to the observed data through a sparse mapping. Using a differential equation representation, we bypass usual computational bottlenecks in obtaining MCMC and online algorithms for approximate Bayesian inference. The performance is assessed in simulations and illustrated in a financial application.

Variational inference algorithms provide the most effective framework for largescale training of Bayesian nonparametric models. Stochastic online approaches are promising, but are sensitive to the chosen learning rate and often converge to poor local optima. We present a new algorithm, memoized online variational inference, which scales to very large (yet finite) datasets while avoiding the complexities of stochastic gradient. Our algorithm maintains finite-dimensional sufficient statistics from batches of the full dataset, requiring some additional memory but still scaling to millions of examples. Exploiting nested families of variational bounds for infinite nonparametric models, we develop principled birth and merge moves allowing non-local optimization. Births adaptively add components to the model to escape local optima, while merges remove redundancy and improve speed. Using Dirichlet process mixture models for image clustering and denoising, we demonstrate major improvements in robustness and accuracy.

Learning a visual concept from a small number of positive examples is a significant challenge for machine learning algorithms. Current methods typically fail to find the appropriate level of generalization in a concept hierarchy for a given set of visual examples. Recent work in cognitive science on Bayesian models of generalization addresses this challenge, but prior results assumed that objects were perfectly recognized. We present an algorithm for learning visual concepts directly from images, using probabilistic predictions generated by visual classifiers as the input to a Bayesian generalization model. As no existing challenge data tests this paradigm, we collect and make available a new, large-scale dataset for visual concept learning using the ImageNet hierarchy as the source of possible concepts, with human annotators to provide ground truth labels as to whether a new image is an instance of each concept using a paradigm similar to that used in experiments studying word learning in children. We compare the performance of our system to several baseline algorithms, and show a significant advantage results from combining visual classifiers with the ability to identify an appropriate level of abstraction using Bayesian generalization.