This work characterizes the benefits of averaging techniques widely used in conjunction with stochastic gradient descent (SGD). In particular, this work sharply analyzes: (1) mini-batching, a method of averaging many samples of the gradient to both reduce the variance of a stochastic gradient estimate and for parallelizing SGD and (2) tail-averaging, a method involving averaging the final few iterates of SGD in order to decrease the variance in SGD's final iterate. This work presents the first tight non-asymptotic generalization error bounds for these schemes for the stochastic approximation problem of least squares regression. Furthermore, this work establishes a precise problem-dependent extent to which mini-batching can be used to yield provable near-linear parallelization speedups over SGD with batch size one. These results are utilized in providing a highly parallelizable SGD algorithm that obtains the optimal statistical error rate with nearly the same number of serial updates as batch gradient descent, which improves significantly over existing SGD-style methods. Finally, this work sheds light on some fundamental differences in SGD's behavior when dealing with agnostic noise in the (non-realizable) least squares regression problem. In particular, the work shows that the stepsizes that ensure optimal statistical error rates for the agnostic case must be a function of the noise properties. The central analysis tools used by this paper are obtained through generalizing the operator view of averaged SGD, introduced by Defossez and Bach (2015) followed by developing a novel analysis in bounding these operators to characterize the generalization error. These techniques may be of broader interest in analyzing various computational aspects of stochastic approximation.

We propose a penalized likelihood method to fit the linear discriminant analysis model when the predictor is matrix valued. We simultaneously estimate the means and the precision matrix, which we assume has a Kronecker product decomposition. Our penalties encourage pairs of response category mean matrices to have equal entries and also encourage zeros in the precision matrix. To compute our estimators, we use a blockwise coordinate descent algorithm. To update the optimization variables corresponding to response category mean matrices, we use an alternating minimization algorithm that takes advantage of the Kronecker structure of the precision matrix. We show that our method can outperform relevant competitors in classification, even when our modeling assumptions are violated. We analyze an EEG dataset to demonstrate our method's interpretability and classification accuracy.

Massively parallel architectures such as the GPU are becoming increasingly important due to the recent proliferation of data. In this paper, we propose a key class of hybrid parallel graphlet algorithms that leverages multiple CPUs and GPUs simultaneously for computing k-vertex induced subgraph statistics (called graphlets). In addition to the hybrid multi-core CPU-GPU framework, we also investigate single GPU methods (using multiple cores) and multi-GPU methods that leverage all available GPUs simultaneously for computing induced subgraph statistics. Both methods leverage GPU devices only, whereas the hybrid multi-core CPU-GPU framework leverages all available multi-core CPUs and multiple GPUs for computing graphlets in large networks. Compared to recent approaches, our methods are orders of magnitude faster, while also more cost effective enjoying superior performance per capita and per watt. In particular, the methods are up to 300 times faster than the recent state-of-the-art method. To the best of our knowledge, this is the first work to leverage multiple CPUs and GPUs simultaneously for computing induced subgraph statistics.

We study the problem of treatment effect estimation in randomized experiments with high-dimensional covariate information, and show that essentially any risk-consistent regression adjustment can be used to obtain efficient estimates of the average treatment effect. Our results considerably extend the range of settings where high-dimensional regression adjustments are guaranteed to provide valid inference about the population average treatment effect. We then propose cross-estimation, a simple method for obtaining finite-sample-unbiased treatment effect estimates that leverages high-dimensional regression adjustments. Our method can be used when the regression model is estimated using the lasso, the elastic net, subset selection, etc. Finally, we extend our analysis to allow for adaptive specification search via cross-validation, and flexible non-parametric regression adjustments with machine learning methods such as random forests or neural networks.

Motivated by online recommendation and advertising systems, we consider a causal model for stochastic contextual bandits with a latent low-dimensional confounder. In our model, there are $L$ observed contexts and $K$ arms of the bandit. The observed context influences the reward obtained through a latent confounder variable with cardinality $m$ ($m \ll L,K$). The arm choice and the latent confounder causally determines the reward while the observed context is correlated with the confounder. Under this model, the $L \times K$ mean reward matrix $\mathbf{U}$ (for each context in $[L]$ and each arm in $[K]$) factorizes into non-negative factors $\mathbf{A}$ ($L \times m$) and $\mathbf{W}$ ($m \times K$). This insight enables us to propose an $\epsilon$-greedy NMF-Bandit algorithm that designs a sequence of interventions (selecting specific arms), that achieves a balance between learning this low-dimensional structure and selecting the best arm to minimize regret. Our algorithm achieves a regret of $\mathcal{O}\left(L\mathrm{poly}(m, \log K) \log T \right)$ at time $T$, as compared to $\mathcal{O}(LK\log T)$ for conventional contextual bandits, assuming a constant gap between the best arm and the rest for each context. These guarantees are obtained under mild sufficiency conditions on the factors that are weaker versions of the well-known Statistical RIP condition. We further propose a class of generative models that satisfy our sufficient conditions, and derive a lower bound of $\mathcal{O}\left(Km\log T\right)$. These are the first regret guarantees for online matrix completion with bandit feedback, when the rank is greater than one. We further compare the performance of our algorithm with the state of the art, on synthetic and real world data-sets.

We provide tight upper and lower bounds on the complexity of minimizing the average of $m$ convex functions using gradient and prox oracles of the component functions. We show a significant gap between the complexity of deterministic vs randomized optimization. For smooth functions, we show that accelerated gradient descent (AGD) and an accelerated variant of SVRG are optimal in the deterministic and randomized settings respectively, and that a gradient oracle is sufficient for the optimal rate. For non-smooth functions, having access to prox oracles reduces the complexity and we present optimal methods based on smoothing that improve over methods using just gradient accesses.

In this paper, we propose a Double Thompson Sampling (D-TS) algorithm for dueling bandit problems. As indicated by its name, D-TS selects both the first and the second candidates according to Thompson Sampling. Specifically, D-TS maintains a posterior distribution for the preference matrix, and chooses the pair of arms for comparison by sampling twice from the posterior distribution. This simple algorithm applies to general Copeland dueling bandits, including Condorcet dueling bandits as its special case. For general Copeland dueling bandits, we show that D-TS achieves $O(K^2 \log T)$ regret. For Condorcet dueling bandits, we further simplify the D-TS algorithm and show that the simplified D-TS algorithm achieves $O(K \log T + K^2 \log \log T)$ regret. Simulation results based on both synthetic and real-world data demonstrate the efficiency of the proposed D-TS algorithm.

Matrix concentration inequalities control the deviation of a random matrix around its mean. Until now, results in literature consider the case of sums of independent random matrices, or matrix martingales in discrete time. A first matrix version of the Chernoff bound is given in [1] and was adapted to yield matrix analogues of standard scalar concentration inequalities in [7, 25, 26, 23]. Later, these results were improved in [34, 33], by the use of a theorem due to Lieb [18], about the concavity of a trace exponential function, which is a deep result closely related to the joint convexity of quantum entropy in physics. See also [20] for a family of sharper and more general results based on the Stein's method. These works contain extensions to random matrices of classical concentration inequalities for sums of independent scalar random variables, such as the Bernstein inequality for sub-exponential random variables, Hoeffding inequality for sub-Gaussian random variables or Freedman inequality for martingales, see e.g.

The text is supported by a strong outline. The authors preserve much of the introductory material, but add the latest techniques and developments in data mining, thus making this a comprehensive resource for both beginners and practitioners. The focus is data-all aspects. The presentation is broad, encyclopedic, and comprehensive, with ample references for interested readers to pursue in-depth research on any technique. "This interesting and comprehensive introduction to data mining emphasizes the interest in multidimensional data mining--the integration of online analytical processing (OLAP) and data mining. Some chapters cover basic methods, and others focus on advanced techniques. The structure, along with the didactic presentation, makes the book suitable for both beginners and specialized readers."

SAN FRANCISCO - 26 Oct 2016: IBM (NYSE: IBM) and Slack are partnering to bring Watson to Slack's global community of developers and enterprise users. Drawing on the power of Slack's digital workplace and the cognitive computing capabilities of Watson, developers will be able to create more offerings -- including bots and other conversational inferences -- that will transform the platform's user experience. Developers can easily access the range of Watson services -- such as Conversation, Sentiment Analysis or speech APIs -- and build powerful new tools for the platform with this enhanced cognitive functionality. As a first step, IBM and Slack intend to develop new and improved communications tools for users of the Slack platform, including an updated Slackbot to be powered by Watson and an IBM Watson-enabled bot for IT and network operations. IBM and Slack also plan to share learnings from the creation of these tools with developers as part of ongoing educational efforts, and the companies plan to offer specialized tutorial resources to accelerate how developers can tap into Watson: Slack Intends to Adopt Watson Conversation for Slackbot: To further strengthen the Slack user experience, Slack intends to adopt Watson Conversation as a technology that helps power its Slackbot -- the platform's popular customer service bot.