Optimal Transport (OT) naturally arises in many machine learning applications, yet the heavy computational burden limits its wide-spread uses. To address the scalability issue, we propose an implicit generative learning-based framework called SPOT (Scalable Push-forward of Optimal Transport). Specifically, we approximate the optimal transport plan by a pushforward of a reference distribution, and cast the optimal transport problem into a minimax problem. We then can solve OT problems efficiently using primal dual stochastic gradient-type algorithms. We also show that we can recover the density of the optimal transport plan using neural ordinary differential equations. Numerical experiments on both synthetic and real datasets illustrate that SPOT is robust and has favorable convergence behavior. SPOT also allows us to efficiently sample from the optimal transport plan, which benefits downstream applications such as domain adaptation.

AdaNet is a lightweight TensorFlow-based (Abadi et al., 2015) framework for automatically learning high-quality ensembles with minimal expert intervention. Our framework is inspired by the AdaNet algorithm (Cortes et al., 2017) which learns the structure of a neural network as an ensemble of subnetworks. We designed it to: (1) integrate with the existing TensorFlow ecosystem, (2) offer sensible default search spaces to perform well on novel datasets, (3) present a flexible API to utilize expert information when available, and (4) efficiently accelerate training with distributed CPU, GPU, and TPU hardware.

Personal use of this material is permitted. Abstract--Given the recent surge in developments of deep x learning, this article provides a review of the state-of-the-art input sequence deep learning techniques for audio signal processing. Subsequently, prominent deep learning application areas are covered, i.e. audio recognition (automatic The number of labels to be predicted (left), and the type of each label (right). While many deep learning methods have been adopted from I. INTRODUCTION Audio [2] in 1986, and finally 3) the success of deep learning in signals are commonly transformed into two-dimensional timefrequency speech recognition [3] and image classification [4] in 2012, representations for processing, but the two axes, leading to a renaissance of deep learning, involving e.g. Images are instantaneous snapshots networks (CNNs, [6]) and long short-term memory (LSTM, of a target and often analyzed as a whole or in patches [7]). In this "deep" paradigm, architectures with a large number with little order constraints; however audio signals have to be of parameters are trained to learn from a massive amount of studied sequentially in chronological order. METHODS many areas of signal processing, often outperforming traditional To set the stage, we give a conceptual overview of audio signal processing on a large scale. In this most recent analysis and synthesis problems (II-A), the input representations wave, deep learning first gained traction in image processing commonly used to address them (II-B), and the models [4], but was then widely adopted in speech processing, music shared between different application fields (II-C). H. Purwins is with Department of Architecture, Design & Media Technology, This division encompasses two independent axes (cf. Manuscript received October 11, 2018 While the audio signal will often be processed into a sequence of features, This is a PREPRINT we consider this part of the solution, not of the task. JOURNAL OF SELECTED TOPICS OF SIGNAL PROCESSING, VOL.

Existing popular methods for semi-supervised learning with Graph Neural Networks (such as the Graph Convolutional Network) provably cannot learn a general class of neighborhood mixing relationships. To address this weakness, we propose a new model, MixHop, that can learn these relationships, including difference operators, by repeatedly mixing feature representations of neighbors at various distances. MixHop requires no additional memory or computational complexity, and outperforms on challenging baselines. In addition, we propose sparsity regularization that allows us to visualize how the network prioritizes neighborhood information across different graph datasets. Our analysis of the learned architectures reveals that neighborhood mixing varies per datasets.

Distributed implementations of gradient-based methods, wherein a server distributes gradient computations across worker machines, suffer from slow running machines, called 'stragglers'. Gradient coding is a coding-theoretic framework to mitigate stragglers by enabling the server to recover the gradient sum in the presence of stragglers. 'Approximate gradient codes' are variants of gradient codes that reduce computation and storage overhead per worker by allowing the server to approximately reconstruct the gradient sum. In this work, our goal is to construct approximate gradient codes that are resilient to stragglers selected by a computationally unbounded adversary. Our motivation for constructing codes to mitigate adversarial stragglers stems from the challenge of tackling stragglers in massive-scale elastic and serverless systems, wherein it is difficult to statistically model stragglers. Towards this end, we propose a class of approximate gradient codes based on balanced incomplete block designs (BIBDs). We show that the approximation error for these codes depends only on the number of stragglers, and thus, adversarial straggler selection has no advantage over random selection. In addition, the proposed codes admit computationally efficient decoding at the server. Next, to characterize fundamental limits of adversarial straggling, we consider the notion of 'adversarial threshold' -- the smallest number of workers that an adversary must straggle to inflict certain approximation error. We compute a lower bound on the adversarial threshold, and show that codes based on symmetric BIBDs maximize this lower bound among a wide class of codes, making them excellent candidates for mitigating adversarial stragglers.

In this paper we seek methods to effectively detect urban micro-events. Urban micro-events are events which occur in cities, have limited geographical coverage and typically affect only a small group of citizens. Because of their scale these are difficult to identify in most data sources. However, by using citizen sensing to gather data, detecting them becomes feasible. The data gathered by citizen sensing is often multimodal and, as a consequence, the information required to detect urban micro-events is distributed over multiple modalities. This makes it essential to have a classifier capable of combining them. In this paper we explore several methods of creating such a classifier, including early, late, hybrid fusion and representation learning using multimodal graphs. We evaluate performance on a real world dataset obtained from a live citizen reporting system. We show that a multimodal approach yields higher performance than unimodal alternatives. Furthermore, we demonstrate that our hybrid combination of early and late fusion with multimodal embeddings performs best in classification of urban micro-events.

Fairness has become a central issue for our research community as classification algorithms are adopted in societally critical domains such as recidivism prediction and loan approval. In this work, we consider the potential bias based on protected attributes (e.g., race and gender), and tackle this problem by learning latent representations of individuals that are statistically indistinguishable between protected groups while sufficiently preserving other information for classification. To do that, we develop a minimax adversarial framework with a generator to capture the data distribution and generate latent representations, and a critic to ensure that the distributions across different protected groups are similar. Our framework provides a theoretical guarantee with respect to statistical parity and individual fairness. Empirical results on four real-world datasets also show that the learned representation can effectively be used for classification tasks such as credit risk prediction while obstructing information related to protected groups, especially when removing protected attributes is not sufficient for fair classification.

When optimizing over-parameterized models, such as deep neural networks, a large set of parameters leads to a zero training error. However they lead to different values for the test error and thus have distinct generalization properties. More specifically, Neyshabur [2017, Part II] argues that the choice of the optimization algorithm (and its respective hyperparameters) provides an implicit regularization with respect to its geometry: it biases the training, finding a particular minimizer of the objective. In this work, we use the same setting as Saxe et al. [2018]: a regression problem with least-square loss on a multidimensional output. Our prediction is made either by a linear model or by a two-layer linear neural network [Saxe et al., 2018]. Our goal is to extend their work on the continuous gradient dynamics in order to understand the behavior of the discrete dynamics induced by these two models. We show that with a vanishing initialization and a small enough step-size, the gradient dynamics of the two-layer linear neural network sequentially learns components that can be ranked according to a hierarchical structure whereas the gradient dynamics of the linear model learns the same components at the same time, missing this notion of hierarchy between components. The path followed by the two-layer formulation actually corresponds to successively solving the initial regression problem with a growing low rank constraint which is also know as reduced-rank regression [Izenman, 1975]. Note that this notion of path followed by the dynamics of a whole network is different from the notion of path introduced by Neyshabur et al. [2015a] which

Sensors which use electromagnetic induction (EMI) to excite a response in conducting bodies have long been investigated for subsurface explosive hazard detection. In particular, EMI sensors have been used to discriminate between different types of objects, and to detect objects with low metal content. One successful, previously investigated approach is the Multiple Instance Adaptive Cosine Estimator (MI-ACE). In this paper, a number of new initialization techniques for MI-ACE are proposed and evaluated using their respective performance and speed. The cross validated learned signatures, as well as learned background statistics, are used with Adaptive Cosine Estimator (ACE) to generate confidence maps, which are clustered into alarms. Alarms are scored against a ground truth and the initialization approaches are compared.

Testing of deep learning models is challenging due to the excessive number and complexity of computations involved. As a result, test data selection is performed manually and in an ad hoc way. This raises the question of how we can automatically select candidate test data to test deep learning models. Recent research has focused on adapting test selection metrics from code-based software testing (such as coverage) to deep learning. However, deep learning models have different attributes from code such as spread of computations across the entire network reflecting training data properties, balance of neuron weights and redundancy (use of many more neurons than needed). Such differences make code-based metrics inappropriate to select data that can challenge the models (can trigger misclassification). We thus propose a set of test selection metrics based on the notion of model uncertainty (model confidence on specific inputs). Intuitively, the more uncertain we are about a candidate sample, the more likely it is that this sample triggers a misclassification. Similarly, the samples for which we are the most uncertain, are the most informative and should be used to improve the model by retraining. We evaluate these metrics on two widely-used image classification problems involving real and artificial (adversarial) data. We show that uncertainty-based metrics have a strong ability to select data that are misclassified and lead to major improvement in classification accuracy during retraining: up to 80% more gain than random selection and other state-of-the-art metrics on one dataset and up to 29% on the other.