We study sampling as optimization in the space of measures. We focus on gradient flow-based optimization with the Langevin dynamics as a case study. We investigate the source of the bias of the unadjusted Langevin algorithm (ULA) in discrete time, and consider how to remove or reduce the bias. We point out the difficulty is that the heat flow is exactly solvable, but neither its forward nor backward method is implementable in general, except for Gaussian data. We propose the symmetrized Langevin algorithm (SLA), which should have a smaller bias than ULA, at the price of implementing a proximal gradient step in space. We show SLA is in fact consistent for Gaussian target measure, whereas ULA is not. We also illustrate various algorithms explicitly for Gaussian target measure, including gradient descent, proximal gradient, and Forward-Backward, and show they are all consistent.

We propose several ways of reusing subword embeddings and other weights in subword-aware neural language models. The proposed techniques do not benefit a competitive character-aware model, but some of them improve the performance of syllable- and morpheme-aware models while showing significant reductions in model sizes. We discover a simple hands-on principle: in a multi-layer input embedding model, layers should be tied consecutively bottom-up if reused at output. Our best morpheme-aware model with properly reused weights beats the competitive word-level model by a large margin across multiple languages and has 20%-87% fewer parameters.

Asynchronous distributed machine learning solutions have proven very effective so far, but always assuming perfectly functioning workers. In practice, some of the workers can however exhibit Byzantine behavior, caused by hardware failures, software bugs, corrupt data, or even malicious attacks. We introduce \emph{Kardam}, the first distributed asynchronous stochastic gradient descent (SGD) algorithm that copes with Byzantine workers. Kardam consists of two complementary components: a filtering and a dampening component. The first is scalar-based and ensures resilience against $\frac{1}{3}$ Byzantine workers. Essentially, this filter leverages the Lipschitzness of cost functions and acts as a self-stabilizer against Byzantine workers that would attempt to corrupt the progress of SGD. The dampening component bounds the convergence rate by adjusting to stale information through a generic gradient weighting scheme. We prove that Kardam guarantees almost sure convergence in the presence of asynchrony and Byzantine behavior, and we derive its convergence rate. We evaluate Kardam on the CIFAR-100 and EMNIST datasets and measure its overhead with respect to non Byzantine-resilient solutions. We empirically show that Kardam does not introduce additional noise to the learning procedure but does induce a slowdown (the cost of Byzantine resilience) that we both theoretically and empirically show to be less than $f/n$, where $f$ is the number of Byzantine failures tolerated and $n$ the total number of workers. Interestingly, we also empirically observe that the dampening component is interesting in its own right for it enables to build an SGD algorithm that outperforms alternative staleness-aware asynchronous competitors in environments with honest workers.

We study the bias of generic optimization methods, including Mirror Descent, Natural Gradient Descent and Steepest Descent with respect to different potentials and norms, when optimizing underdetermined linear regression or separable linear classification problems. We ask the question of whether the global minimum (among the many possible global minima) reached by optimization algorithms can be characterized in terms of the potential or norm, and independently of hyperparameter choices such as step size and momentum.

The $k$-means algorithm is the most popular nonparametric clustering method in use, but cannot generally be applied to data sets with missing observations. The usual practice with such data sets is to either impute the values under an assumption of a missing-at-random mechanism or to ignore the incomplete records, and then to use the desired clustering method. We develop an efficient version of the $k$-means algorithm that allows for clustering cases where not all the features have observations recorded. Our extension is called $k_m$-means and reduces to the $k$-means algorithm when all records are complete. We also provide strategies to initialize our algorithm and to estimate the number of groups in the data set. Illustrations and simulations demonstrate the efficacy of our approach in a variety of settings and patterns of missing data. Our methods are also applied to the clustering of gamma-ray bursts and to the analysis of activation images obtained from a functional Magnetic Resonance Imaging experiment.

In this paper we consider the low-rank matrix completion problem with specific application to forecasting in time series analysis. Briefly, the low-rank matrix completion problem is the problem of imputing missing values of a matrix under a rank constraint. We consider a matrix completion problem for Hankel matrices and a convex relaxation based on the nuclear norm. Based on new theoretical results and a number of numerical and real examples, we investigate the cases when the proposed approach can work. Our results highlight the importance of choosing a proper weighting scheme for the known observations.

One of the main drivers behind the rapid recent advances in machine learning has been the availability of efficient system support. This comes both through hardware acceleration, but also in the form of efficient software frameworks and programming models. Despite significant progress, scaling compute-intensive machine learning workloads to a large number of compute nodes is still a challenging task, with significant latency and bandwidth demands. In this paper, we address this challenge, by proposing SPARCML, a general, scalable communication layer for machine learning applications. SPARCML is built on the observation that many distributed machine learning algorithms either have naturally sparse communication patters, or have updates which can be sparsified in a structured way for improved performance, without any convergence or accuracy loss. To exploit this insight, we design and implement a set of communication efficient protocols for sparse input data, in conjunction with efficient machine learning algorithms which can leverage these primitives. Our communication protocols generalize standard collective operations, by allowing processes to contribute sparse input data vectors, of heterogeneous sizes. We call these operations sparse-input collectives, and present efficient practical algorithms with strong theoretical bounds on their running time and communication cost. Our generic communication layer is enriched with additional features, such support for non-blocking (asynchronous) operations, and support for low-precision data representations. We validate our algorithmic results experimentally on a range of large-scale machine learning applications and target architectures, showing that we can leverage sparsity for order- of-magnitude runtime savings, compared to state-of-the art methods and frameworks.

Topic models are one of the most frequently used models in machine learning due to its high interpretability and modular structure. However extending the model to include supervisory signal, incorporate pre-trained word embedding vectors and add nonlinear output function to the model is not an easy task because one has to resort to highly intricate approximate inference procedure. In this paper, we show that topic models could be viewed as performing a neighborhood aggregation algorithm where the messages are passed through a network defined over words. Under the network view of topic models, nodes corresponds to words in a document and edges correspond to either a relationship describing co-occurring words in a document or a relationship describing same word in the corpus. The network view allows us to extend the model to include supervisory signals, incorporate pre-trained word embedding vectors and add nonlinear output function to the model in a simple manner. Moreover, we describe a simple way to train the model that is well suited in a semi-supervised setting where we only have supervisory signals for some portion of the corpus and the goal is to improve prediction performance in the held-out data. Through careful experiments we show that our approach outperforms state-of-the-art supervised Latent Dirichlet Allocation implementation in both held-out document classification tasks and topic coherence.

We propose and analyze a variant of the classic Polyak-Ruppert averaging scheme, broadly used in stochastic gradient methods. Rather than a uniform average of the iterates, we consider a weighted average, with weights decaying in a geometric fashion. In the context of linear least squares regression, we show that this averaging scheme has a the same regularizing effect, and indeed is asymptotically equivalent, to ridge regression. In particular, we derive finite-sample bounds for the proposed approach that match the best known results for regularized stochastic gradient methods.

We study the robustness of classifiers to various kinds of random noise models. In particular, we consider noise drawn uniformly from the $\ell\_p$ ball for $p \in [1, \infty]$ and Gaussian noise with an arbitrary covariance matrix. We characterize this robustness to random noise in terms of the distance to the decision boundary of the classifier. This analysis applies to linear classifiers as well as classifiers with locally approximately flat decision boundaries, a condition which is satisfied by state-of-the-art deep neural networks. The predicted robustness is verified experimentally.