We introduce Generative Neural Machine Translation (GNMT), a latent variable architecture which is designed to model the semantics of the source and target sentences. We modify an encoder-decoder translation model by adding a latent variable as a language agnostic representation which is encouraged to learn the meaning of the sentence. GNMT achieves competitive BLEU scores on pure translation tasks, and is superior when there are missing words in the source sentence. We augment the model to facilitate multilingual translation and semi-supervised learning without adding parameters. This framework significantly reduces overfitting when there is limited paired data available, and is effective for translating between pairs of languages not seen during training.

Abstract--In a world of global trading, maritime safety, security and efficiency are crucial issues. We propose a multi-task deep learning framework for vessel monitoring using Automatic Identification System (AIS) data streams. We combine recurrent neural networks with latent variable modeling and an embedding of AIS messages to a new representation space to jointly address key issues to be dealt with when considering AIS data streams: massive amount of streaming data, noisy data and irregular timesampling. We demonstrate the relevance of the proposed deep learning framework on real AIS datasets for a three-task setting, namely trajectory reconstruction, anomaly detection and vessel type identification. In the world of a globalized economy, maritime surveillance is a vital demand. Besides, the real-time delivery of maritime situation maps is also necessary for a variety of activities: fishing activities control, smuggling detection, EEZ intrusion detection, transshipment detection, maritime pollution monitoring, etc. Over the last decades, the development of terrestrial networks and satellite constellations of Automatic Identification System (AIS) has opened a new era in maritime traffic surveillance. Every day, AIS provides on a global scale hundreds of millions of messages [1], which contain ships' identifiers, their Global Positioning System (GPS) coordinates, their speed, course, etc. The potential of this massive amount of data is clearly of interest if tools and models provide means to efficiently extract, detect and analyze relevant information from these data streams. However, current operational systems, which strongly rely on human experts, can only deal with a limited fraction of AIS data streams. Thus, the development of AIbased systems is a critical challenge.

DNA is perceived as a sequence over the letters {A, C, G, T }, the alphabet of nucleotides. This sequence constitutes the code that acts as a blueprint for all processes taking place in a cell. But beyond merely reflecting primary sequence, DNA is a molecule, which implies that DNA assumes spatial structure and shape. The spatial organization of DNA is achieved by integrating ("recruiting") other molecules, the histone proteins, that help to assume the correct spatial configuration. The combination of DNA and helper molecules is called chromatin; the spatial configuration of the chromatin, finally, defines the functional properties of local areas of the DNA [de Graaf and van Steensel, 2013]. Chromatin can assume several function-defining epigenetic states, where states vary along the genome [Ernst et al., 2011]. The key determinant for spatial configuration is the underlying primary DNA sequence: sequential patterns are responsible for recruiting histone proteins and their chemical modifications, which in turn give rise to or even define the chromatin states. The exact configuration of the chromatin and its interplay with the underlying raw DNA sequence are under active research.

Our paper proposes a generalization error bound for a general family of deep neural networks based on the spectral norm of weight matrices. Through introducing a novel characterization of the Lipschitz properties of neural network family, we achieve a tighter generalization error bound for ultra-deep neural networks, whose depth is much larger than the square root of its width. Besides the general deep neural networks, our results can be applied to derive new bounds for several popular architectures, including convolutional neural networks (CNNs), residual networks (ResNets), and hyperspherical networks (SphereNets). In the regime that the depth of these architectures is dominating, our bounds allow for the choice of much larger parameter spaces of weight matrices, inducing potentially stronger expressive ability.

Deep neural networks, and in particular recurrent networks, are promising candidates to control autonomous agents that interact in real-time with the physical world. However, this requires a seamless integration of temporal features into the network's architecture. For the training of and inference with recurrent neural networks, they are usually rolled out over time, and different rollouts exist. Conventionally, during inference the layers of a network are computed in a sequential manner resulting in sparse temporal integration of information and long response times. In this study, we present a theoretical framework to describe the set of all rollouts and demonstrate their differences in solving specific tasks. We prove that certain rollouts, also with only skip and no recurrent connections, enable earlier and more frequent responses, and show empirically that these early responses have better performance. The streaming rollout maximizes these properties and, in addition, enables a fully parallel execution of the network reducing the runtime on massively parallel devices. Additionally, we provide an open-source toolbox to design, train, evaluate, and online-interact with streaming rollouts.

One of the current issues in Brain-Computer Interface is how to deal with noisy Electroencephalography measurements organized as multidimensional datasets. On the other hand, recently, significant advances have been made in multidimensional signal completion algorithms that exploit tensor decomposition models to capture the intricate relationship among entries in a multidimensional signal. We propose to use tensor completion applied to EEG data for improving the classification performance in a motor imagery BCI system with corrupted measurements. Noisy measurements are considered as unknowns that are inferred from a tensor decomposition model. We evaluate the performance of four recently proposed tensor completion algorithms plus a simple interpolation strategy, first with random missing entries and then with missing samples constrained to have a specific structure (random missing channels), which is a more realistic assumption in BCI Applications. We measured the ability of these algorithms to reconstruct the tensor from observed data. Then, we tested the classification accuracy of imagined movement in a BCI experiment with missing samples. We show that for random missing entries, all tensor completion algorithms can recover missing samples increasing the classification performance compared to a simple interpolation approach. For the random missing channels case, we show that tensor completion algorithms help to reconstruct missing channels, significantly improving the accuracy in the classification of motor imagery, however, not at the same level as clean data. Tensor completion algorithms are useful in real BCI applications. The proposed strategy could allow using motor imagery BCI systems even when EEG data is highly affected by missing channels and/or samples, avoiding the need of new acquisitions in the calibration stage.

In (Franceschi et al., 2018) we proposed a unified mathematical framework, grounded on bilevel programming, that encompasses gradient-based hyperparameter optimization and meta-learning. We formulated an approximate version of the problem where the inner objective is solved iteratively, and gave sufficient conditions ensuring convergence to the exact problem. In this work we show how to optimize learning rates, automatically weight the loss of single examples and learn hyper-representations with Far-HO, a software package based on the popular deep learning framework TensorFlow that allows to seamlessly tackle both HO and ML problems.

Many real-world vision problems suffer from inherent ambiguities. In clinical applications for example, it might not be clear from a CT scan alone which particular region is cancer tissue. Therefore a group of graders typically produces a set of diverse but plausible segmentations. We consider the task of learning a distribution over segmentations given an input. To this end we propose a generative segmentation model based on a combination of a U-Net with a conditional variational autoencoder that is capable of efficiently producing an unlimited number of plausible hypotheses. We show on a lung abnormalities segmentation task and on a Cityscapes segmentation task that our model reproduces the possible segmentation variants as well as the frequencies with which they occur, doing so significantly better than published approaches. These models could have a high impact in real-world applications, such as being used as clinical decision-making algorithms accounting for multiple plausible semantic segmentation hypotheses to provide possible diagnoses and recommend further actions to resolve the present ambiguities.

A good classification method should yield more accurate results than simple heuristics. But there are classification problems, especially high-dimensional ones like the ones based on image/video data, for which simple heuristics can work quite accurately; the structure of the data in such problems is easy to uncover without any sophisticated or computationally expensive method. On the other hand, some problems have a structure that can only be found with sophisticated pattern recognition methods. We are interested in quantifying the difficulty of a given high-dimensional pattern recognition problem. We consider the case where the patterns come from two pre-determined classes and where the objects are represented by points in a high-dimensional vector space. However, the framework we propose is extendable to an arbitrarily large number of classes. We propose classification benchmarks based on simple random projection heuristics. Our benchmarks are 2D curves parameterized by the classification error and computational cost of these simple heuristics. Each curve divides the plane into a "positive- gain" and a "negative-gain" region. The latter contains methods that are ill-suited for the given classification problem. The former is divided into two by the curve asymptote; methods that lie in the small region under the curve but right of the asymptote merely provide a computational gain but no structural advantage over the random heuristics. We prove that the curve asymptotes are optimal (i.e. at Bayes error) in some cases, and thus no sophisticated method can provide a structural advantage over the random heuristics. Such classification problems, an example of which we present in our numerical experiments, provide poor ground for testing new pattern classification methods.

We explore the use of deep learning and deep reinforcement learning for optimization problems in transportation. Many transportation system analysis tasks are formulated as an optimization problem - such as optimal control problems in intelligent transportation systems and long term urban planning. Often transportation models used to represent dynamics of a transportation system involve large data sets with complex input-output interactions and are difficult to use in the context of optimization. Use of deep learning metamodels can produce a lower dimensional representation of those relations and allow to implement optimization and reinforcement learning algorithms in an efficient manner. In particular, we develop deep learning models for calibrating transportation simulators and for reinforcement learning to solve the problem of optimal scheduling of travelers on the network.