Canonical Correlation Analysis (CCA) is a linear representation learning method that seeks maximally correlated variables in multi-view data. Non-linear CCA extends this notion to a broader family of transformations, which are more powerful for many real-world applications. Given the joint probability, the Alternating Conditional Expectation (ACE) provides an optimal solution to the non-linear CCA problem. However, it suffers from limited performance and an increasing computational burden when only a finite number of observations is available. In this work we introduce an information-theoretic framework for the non-linear CCA problem (ITCCA), which extends the classical ACE approach. Our suggested framework seeks compressed representations of the data that allow a maximal level of correlation. This way we control the trade-off between the flexibility and the complexity of the representation. Our approach demonstrates favorable performance at a reduced computational burden, compared to non-linear alternatives, in a finite sample size regime. Further, ITCCA provides theoretical bounds and optimality conditions, as we establish fundamental connections to rate-distortion theory, the information bottleneck and remote source coding. In addition, it implies a "soft" dimensionality reduction, as the compression level is measured (and governed) by the mutual information between the original noisy data and the signals that we extract.

Leverage score sampling provides an appealing way to perform approximate computations for large matrices. Indeed, it allows to derive faithful approximations with a complexity adapted to the problem at hand. Yet, performing leverage scores sampling is a challenge in its own right requiring further approximations. In this paper, we study the problem of leverage score sampling for positive definite matrices defined by a kernel. Our contribution is twofold. First we provide a novel algorithm for leverage score sampling and second, we exploit the proposed method in statistical learning by deriving a novel solver for kernel ridge regression. Our main technical contribution is showing that the proposed algorithms are currently the most efficient and accurate for these problems.

We propose a practical approach based on federated learning to solve out-of-domain issues with continuously running embedded speech-based models such as wake word detectors. We conduct an extensive empirical study of the federated averaging algorithm for the "Hey Snips" wake word based on a crowdsourced dataset that mimics a federation of wake word users. We empirically demonstrate that using an adaptive averaging strategy inspired from Adam in place of standard weighted model averaging highly reduces the number of communication rounds required to reach our target performance. The associated upstream communication costs per user are estimated at 8 MB, which is a reasonable in the context of smart home voice assistants. Additionally, the dataset used for these experiments is being open sourced with the aim of fostering further transparent research in the application of federated learning to speech data.

We present a probabilistic modeling framework and adaptive sampling algorithm wherein unsupervised generative models are combined with black box predictive models to tackle the problem of input design. In input design, one is given one or more stochastic "oracle" predictive functions, each of which maps from the input design space (e. g., DNA sequences or images) to a distribution over a property of interest (e. g., protein fluorescence or image content). Given such stochastic oracles, the problem is to find an input that is expected to maximize one or more properties, or to achieve a specified value of one or more properties, or any combination thereof. We demonstrate experimentally that our approach substantially outperforms other recently presented methods for tackling a specific version of this problem, namely, maximization when the oracle is assumed to be deterministic and unbiased. We also demonstrate that our method can tackle more general versions of the problem.

High-dimensional and sparse data are commonly encountered in many applications of machine learning, such as computer vision, bioinformatics, text mining and behavioral targeting. To classify, cluster or rank data points, it is important to be able to compute semantically meaningful similarities between them. However, defining an appropriate similarity measure for a given task is often difficult as only a small and unknown subset of all features are actually relevant. For instance, in drug discovery studies, chemical compounds are typically represented by a large number of sparse features describing their 2D and 3D properties, and only a few of them play in role in determining whether the compound will bind to a particular target receptor (Leach and Gillet, 2007). In text classification and clustering, a document is often represented as a sparse bag of words, and only a small subset of the dictionary is generally useful to discriminate between documents about different topics. Another example is targeted advertising, where ads are selected based on fine-grained user history (Chen et al., 2009). Similarity and metric learning (Bellet et al., 2015) offers principled approaches to construct a taskspecific similarity measure by learning it from weakly supervised data, and has been used in many application domains. The main theme in these methods is to learn the parameters of a similarity (or distance) function such that it agrees with task-specific similarity judgments (e.g., of the form "data point x should

The development of a metric for structural data is a long-term problem in pattern recognition and machine learning. In this paper, we develop a general metric for comparing nonlinear dynamical systems that is defined with Perron-Frobenius operators in reproducing kernel Hilbert spaces. Our metric includes the existing fundamental metrics for dynamical systems, which are basically defined with principal angles between some appropriately-chosen subspaces, as its special cases. We also describe the estimation of our metric from finite data. We empirically illustrate our metric with an example of rotation dynamics in a unit disk in a complex plane, and evaluate the performance with real-world time-series data.

Unsupervised and semi-supervised learning are important problems that are especially challenging with complex data like natural images. Progress on these problems would accelerate if we had access to appropriate generative models under which to pose the associated inference tasks. Inspired by the success of Convolutional Neural Networks (CNNs) for supervised prediction in images, we design the Neural Rendering Model (NRM), a new probabilistic generative model whose inference calculations correspond to those in a given CNN architecture. The NRM uses the given CNN to design the prior distribution in the probabilistic model. Furthermore, the NRM generates images from coarse to finer scales. It introduces a small set of latent variables at each level, and enforces dependencies among all the latent variables via a conjugate prior distribution. This conjugate prior yields a new regularizer based on paths rendered in the generative model for training CNNs-the Rendering Path Normalization (RPN). We demonstrate that this regularizer improves generalization, both in theory and in practice. In addition, likelihood estimation in the NRM yields training losses for CNNs, and inspired by this, we design a new loss termed as the Max-Min cross entropy which outperforms the traditional cross-entropy loss for object classification. The Max-Min cross entropy suggests a new deep network architecture, namely the Max-Min network, which can learn from less labeled data while maintaining good prediction performance. Our experiments demonstrate that the NRM with the RPN and the Max-Min architecture exceeds or matches the-state-of-art on benchmarks including SVHN, CIFAR10, and CIFAR100 for semi-supervised and supervised learning tasks.

We investigate how to train kernel approximation methods that generalize well under a memory budget. Building on recent theoretical work, we define a measure of kernel approximation error which we find to be much more predictive of the empirical generalization performance of kernel approximation methods than conventional metrics. An important consequence of this definition is that a kernel approximation matrix must be high-rank to attain close approximation. Because storing a high-rank approximation is memory-intensive, we propose using a low-precision quantization of random Fourier features (LP-RFFs) to build a high-rank approximation under a memory budget. Theoretically, we show quantization has a negligible effect on generalization performance in important settings. Empirically, we demonstrate across four benchmark datasets that LP-RFFs can match the performance of full-precision RFFs and the Nystr\"{o}m method, with 3x-10x and 50x-460x less memory, respectively.

We introduce an improved variational autoencoder (VAE) for text modeling with topic information explicitly modeled as a Dirichlet latent variable. By providing the proposed model topic awareness, it is more superior at reconstructing input texts. Furthermore, due to the inherent interactions between the newly introduced Dirichlet variable and the conventional multivariate Gaussian variable, the model is less prone to KL divergence vanishing. We derive the variational lower bound for the new model and conduct experiments on four different data sets. The results show that the proposed model is superior at text reconstruction across the latent space and classifications on learned representations have higher test accuracies.

One of the main challenges in cluster analysis is estimating the true number of clusters in a dataset. This paper quantifies a notion of persistence of a clustering solution over a range of resolution scales, which is used to characterize the natural clusters and estimate the true number of clusters in a dataset. We show that this quantification of persistence is associated with evaluating the largest eigenvalue of the underlying cluster covariance matrix. Detailed experiments on a variety of standard and synthetic datasets demonstrate that the proposed persistence-based indicator outperforms the existing approaches, such as, gap-statistic method, $X$-means, $G$-means, $PG$-means, dip-means algorithms and information-theoretic method, in accurately predicting the true number of clusters. Interestingly, our method can be explained in terms of the phase-transition phenomenon in the deterministic annealing algorithm where the number of cluster centers changes (bifurcates) with respect to an annealing parameter. However, the approach suggested in this paper is independent of the choice of clustering algorithm; and can be used in conjunction with any suitable clustering algorithm.