Cluster algorithms are gaining in popularity due to their compelling ability to identify discrete subgroups in data, and their increasing accessibility in mainstream programming languages and statistical software. While researchers can follow guidelines to choose the right algorithms, and to determine what constitutes convincing clustering, there are no firmly established ways of computing a priori statistical power for cluster analysis. Here, we take a simulation approach to estimate power and classification accuracy for popular analysis pipelines. We systematically varied cluster size, number of clusters, number of different features between clusters, effect size within each different feature, and cluster covariance structure in generated datasets. We then subjected these datasets to common dimensionality reduction approaches (none, multi-dimensional scaling, or uniform manifold approximation and projection) and cluster algorithms (k-means, hierarchical agglomerative clustering with Ward linkage and Euclidean distance, or average linkage and cosine distance, HDBSCAN). Furthermore, we simulated additional datasets to explore the effect of sample size and cluster separation on statistical power and classification accuracy. We found that clustering outcomes were driven by large effect sizes or the accumulation of many smaller effects across features, and were mostly unaffected by differences in covariance structure. Sufficient statistical power can be achieved with relatively small samples (N=20 per subgroup), provided cluster separation is large ({\Delta}=4). Finally, we discuss whether fuzzy clustering (c-means) could provide a more parsimonious alternative for identifying separable multivariate normal distributions, particularly those with lower centroid separation.

Starting with Gilmer et al. (2018), several works have demonstrated the inevitability of adversarial examples based on different assumptions about the underlying input probability space. It remains unclear, however, whether these results apply to natural image distributions. In this work, we assume the underlying data distribution is captured by some conditional generative model, and prove intrinsic robustness bounds for a general class of classifiers, which solves an open problem in Fawzi et al. (2018). Building upon the state-of-the-art conditional generative models, we study the intrinsic robustness of two common image benchmarks under $\ell_2$ perturbations, and show the existence of a large gap between the robustness limits implied by our theory and the adversarial robustness achieved by current state-of-the-art robust models. Code for all our experiments is available at https://github.com/xiaozhanguva/Intrinsic-Rob.

We propose a penalized likelihood framework for estimating multiple precision matrices from different classes. Most existing methods either incorporate no information on relationships between the precision matrices, or require this information be known a priori. The framework proposed in this article allows for simultaneous estimation of the precision matrices and relationships between the precision matrices, jointly. Sparse and non-sparse estimators are proposed, both of which require solving a non-convex optimization problem. To compute our proposed estimators, we use an iterative algorithm which alternates between a convex optimization problem solved by blockwise coordinate descent and a k-means clustering problem. Blockwise updates for computing the sparse estimator require solving an elastic net penalized precision matrix estimation problem, which we solve using a proximal gradient descent algorithm. We prove that this subalgorithm has a linear rate of convergence. In simulation studies and two real data applications, we show that our method can outperform competitors that ignore relevant relationships between precision matrices and performs similarly to methods which use prior information often uknown in practice.

We consider a budget-constrained bandit problem where each arm pull incurs a random cost, and yields a random reward in return. The objective is to maximize the total expected reward under a budget constraint on the total cost. The model is general in the sense that it allows correlated and potentially heavy-tailed cost-reward pairs that can take on negative values as required by many applications. We show that if moments of order $(2+\gamma)$ for some $\gamma > 0$ exist for all cost-reward pairs, $O(\log B)$ regret is achievable for a budget $B>0$. In order to achieve tight regret bounds, we propose algorithms that exploit the correlation between the cost and reward of each arm by extracting the common information via linear minimum mean-square error estimation. We prove a regret lower bound for this problem, and show that the proposed algorithms achieve tight problem-dependent regret bounds, which are optimal up to a universal constant factor in the case of jointly Gaussian cost and reward pairs.

We consider the use of decision trees for decision-making problems under the predict-then-optimize framework. That is, we would like to first use a decision tree to predict unknown input parameters of an optimization problem, and then make decisions by solving the optimization problem using the predicted parameters. A natural loss function in this framework is to measure the suboptimality of the decisions induced by the predicted input parameters, as opposed to measuring loss using input parameter prediction error. This natural loss function is known in the literature as the Smart Predict-then-Optimize (SPO) loss, and we propose a tractable methodology called SPO Trees (SPOTs) for training decision trees under this loss. SPOTs benefit from the interpretability of decision trees, providing an interpretable segmentation of contextual features into groups with distinct optimal solutions to the optimization problem of interest. We conduct several numerical experiments on synthetic and real data including the prediction of travel times for shortest path problems and predicting click probabilities for news article recommendation. We demonstrate on these datasets that SPOTs simultaneously provide higher quality decisions and significantly lower model complexity than other machine learning approaches (e.g., CART) trained to minimize prediction error.

Conventional survival analysis approaches estimate risk scores or individualized time-to-event distributions conditioned on covariates. In practice, there is often great population-level phenotypic heterogeneity, resulting from (unknown) subpopulations with diverse risk profiles or survival distributions. As a result, there is an unmet need in survival analysis for identifying subpopulations with distinct risk profiles, while jointly accounting for accurate individualized time-to-event predictions. An approach that addresses this need is likely to improve characterization of individual outcomes by leveraging regularities in subpopulations, thus accounting for population-level heterogeneity. In this paper, we propose a Bayesian nonparametrics approach that represents observations (subjects) in a clustered latent space, and encourages accurate time-to-event predictions and clusters (subpopulations) with distinct risk profiles. Experiments on real-world datasets show consistent improvements in predictive performance and interpretability relative to existing state-of-the-art survival analysis models.

Reliable uncertainty estimates are an important tool for helping autonomous agents or human decision makers understand and leverage predictive models. However, existing approaches to estimating uncertainty largely ignore the possibility of covariate shift--i.e., where the real-world data distribution may differ from the training distribution. As a consequence, existing algorithms can overestimate certainty, possibly yielding a false sense of confidence in the predictive model. We propose an algorithm for calibrating predictions that accounts for the possibility of covariate shift, given labeled examples from the training distribution and unlabeled examples from the real-world distribution. Our algorithm uses importance weighting to correct for the shift from the training to the real-world distribution. However, importance weighting relies on the training and real-world distributions to be sufficiently close. Building on ideas from domain adaptation, we additionally learn a feature map that tries to equalize these two distributions. In an empirical evaluation, we show that our proposed approach outperforms existing approaches to calibrated prediction when there is covariate shift.

The success of deep learning is due, to a great extent, to the remarkable effectiveness of gradient-based optimization methods applied to large neural networks. In this work we isolate some general mathematical structures allowing for efficient optimization in over-parameterized systems of non-linear equations, a setting that includes deep neural networks. In particular, we show that optimization problems corresponding to such systems are not convex, even locally, but instead satisfy the Polyak-Lojasiewicz (PL) condition allowing for efficient optimization by gradient descent or SGD. We connect the PL condition of these systems to the condition number associated to the tangent kernel and develop a non-linear theory parallel to classical analyses of over-parameterized linear equations. We discuss how these ideas apply to training shallow and deep neural networks. Finally, we point out that tangent kernels associated to certain large system may be far from constant, even locally. Yet, our analysis still allows to demonstrate existence of solutions and convergence of gradient descent and SGD.

Gradient Langevin dynamics (GLD) and stochastic GLD (SGLD) have attracted considerable attention lately, as a way to provide convergence guarantees in a non-convex setting. However, the known rates grow exponentially with the dimension of the space. In this work, we provide a convergence analysis of GLD and SGLD when the optimization space is an infinite dimensional Hilbert space. More precisely, we derive non-asymptotic, dimension-free convergence rates for GLD/SGLD when performing regularized non-convex optimization in a reproducing kernel Hilbert space. Amongst others, the convergence analysis relies on the properties of a stochastic differential equation, its discrete time Galerkin approximation and the geometric ergodicity of the associated Markov chains.

We propose a method to reduce false voice triggers of a speech-enabled personal assistant by post-processing the hypothesis lattice of a server-side large-vocabulary continuous speech recognizer (LVCSR) via a neural network. We first discuss how an estimate of the posterior probability of the trigger phrase can be obtained from the hypothesis lattice using known techniques to perform detection, then investigate a statistical model that processes the lattice in a more explicitly data-driven, discriminative manner. We propose using a Bidirectional Lattice Recurrent Neural Network (LatticeRNN) for the task, and show that it can significantly improve detection accuracy over using the 1-best result or the posterior.