This paper summarizes some recent advances on a set of tasks related to the processing of singing using state-of-the-art deep learning techniques. We discuss their achievements in terms of accuracy and sound quality, and the current challenges, such as availability of data and computing resources. We also discuss the impact that these advances do and will have on listeners and singers when they are integrated in commercial applications.

We introduce a novel generative autoencoder network model that learns to encode and reconstruct images with high quality and resolution, and supports smooth random sampling from the latent space of the encoder. Generative adversarial networks (GANs) are known for their ability to simulate random high-quality images, but they cannot reconstruct existing images. Previous works have attempted to extend GANs to support such inference but, so far, have not delivered satisfactory high-quality results. Instead, we propose the Progressively Growing Generative Autoencoder (PIONEER) network which achieves high-quality reconstruction with $128{\times}128$ images without requiring a GAN discriminator. We merge recent techniques for progressively building up the parts of the network with the recently introduced adversarial encoder-generator network. The ability to reconstruct input images is crucial in many real-world applications, and allows for precise intelligent manipulation of existing images. We show promising results in image synthesis and inference, with state-of-the-art results in CelebA inference tasks.

At multi TeV pp colliders such as the LHC, boosted heavy particles can be produced and form a single collimated cluster of particles. Such a localized cluster is distinguished from QCD jets from quarks or gluons by the substructures of the cluster [1]. For this purpose, consistent definitions of substructures of jets have been studied extensively. There are various methods for identifying the jet substructures, such as strategies based on cluster decomposition [1-8] and shape variables [9-13]. These methods focus on different features of jet substructures to maximize the discrimination power. For the case of Higgs, W, and Z boson decaying hadronically into two quarks, a critical feature is a two-prong substructure inside. Because the key features depend on nature of the parent particle of a jet, there are several frameworks that can be applied to jets [14-18]. In this paper, we propose a new framework to identify jet substructures using a spectral function similar to the angular structure function [14, 19, 20]. Spectral analysis is widely used technique to explore quantum worlds.

We propose an efficient protocol for decentralized training of deep neural networks from distributed data sources. The proposed protocol allows to handle different phases of model training equally well and to quickly adapt to concept drifts. This leads to a reduction of communication by an order of magnitude compared to periodically communicating state-of-the-art approaches. Moreover, we derive a communication bound that scales well with the hardness of the serialized learning problem. The reduction in communication comes at almost no cost, as the predictive performance remains virtually unchanged. Indeed, the proposed protocol retains loss bounds of periodically averaging schemes. An extensive empirical evaluation validates major improvement of the trade-off between model performance and communication which could be beneficial for numerous decentralized learning applications, such as autonomous driving, or voice recognition and image classification on mobile phones.

Collectively, machine learning (ML) researchers are engaged in the creation and dissemination of knowledge about data-driven algorithms. In a given paper, researchers might aspire to any subset of the following goals, among others: to theoretically characterize what is learnable, to obtain understanding through empirically rigorous experiments, or to build a working system that has high predictive accuracy. While determining which knowledge warrants inquiry may be subjective, once the topic is fixed, papers are most valuable to the community when they act in service of the reader, creating foundational knowledge and communicating as clearly as possible. Recent progress in machine learning comes despite frequent departures from these ideals. In this paper, we focus on the following four patterns that appear to us to be trending in ML scholarship: (i) failure to distinguish between explanation and speculation; (ii) failure to identify the sources of empirical gains, e.g., emphasizing unnecessary modifications to neural architectures when gains actually stem from hyper-parameter tuning; (iii) mathiness: the use of mathematics that obfuscates or impresses rather than clarifies, e.g., by confusing technical and non-technical concepts; and (iv) misuse of language, e.g., by choosing terms of art with colloquial connotations or by overloading established technical terms. While the causes behind these patterns are uncertain, possibilities include the rapid expansion of the community, the consequent thinness of the reviewer pool, and the often-misaligned incentives between scholarship and short-term measures of success (e.g., bibliometrics, attention, and entrepreneurial opportunity). While each pattern offers a corresponding remedy (don't do it), we also discuss some speculative suggestions for how the community might combat these trends.

We introduce a new neural network architecture, IGLOO, which aims at providing a representation for long sequences where RNNs fail to converge. The structure uses the relationships between random patches sliced out of the features space of some backbone 1 dimensional CNN to find a representation. This paper explains the implementation of the method and provides benchmark results commonly used for RNNs and compare IGLOO to other structures recently published. It is found that IGLOO can deal with sequences of up to 25,000 time steps. For shorter sequences it is also found to be effective and we find that it achieves the highest score in the literature for the permuted MNIST task. Benchmarks also show that IGLOO can run at the speed of the CuDNN optimised GRU or LSTM without being tied to any specific hardware.

For sufficiently large datasets, neural engines tend to outshine statistical models in most natural language processing problems. However, a speech accent remains a challenge for both approaches. Phonologists manually create general rules describing a speaker's accent, but their results remain underutilized. In this paper, we propose a model that automatically retrieves phonological generalizations from a small dataset. This method leverages the difference in pronunciation between a particular dialect and General American English (GAE) and creates new accented samples of words. The proposed model is able to learn all generalizations that previously were manually obtained by phonologists. We use this statistical method to generate a million phonological variations of words from the CMU Pronouncing Dictionary and train a sequence-to-sequence RNN to recognize accented words with 59% accuracy.

We propose a method to decrease the number of hidden units of the restricted Boltzmann machine while avoiding decrease of the performance measured by the Kullback-Leibler divergence. Then, we demonstrate our algorithm by using numerical simulations.

Tornadoes are the most violent of all atmospheric storms. In a typical year, the United States experiences hundreds of tornadoes with associated damages on the order of one billion dollars. Community preparation and resilience would benefit from accurate predictions of these economic losses, particularly as populations in tornado-prone areas continue to increase in density and extent. Here, we use artificial neural networks to predict tornado-induced property damage using publicly available data. We find that the large number of tornadoes which cause zero property damage (30.6% of the data) poses a challenge for predictive models. We developed a model that predicts whether a tornado will cause property damage to a high degree of accuracy (out of sample accuracy = 0.829 and AUROC = 0.873). Conditional on a tornado causing damage, another model predicts the amount of damage. When combined, these two models yield an expected value for the amount of property damage caused by a tornado event. From the best-performing models (out of sample mean squared error = 0.089 and R2 = 0.473), we provide an interactive, gridded map of monthly expected values for the year 2018. One major weakness is that the model predictive power is optimized with log-transformed, mean-normalized property damages, however this leads to large natural-scale residuals for the most destructive tornadoes. The predictive capacity of this model along with an interactive interface may provide an opportunity for science-informed tornado disaster planning.

The requirement of large amounts of annotated images has become one grand challenge while training deep neural network models for various visual detection and recognition tasks. This paper presents a novel image synthesis technique that aims to generate a large amount of annotated scene text images for training accurate and robust scene text detection and recognition models. The proposed technique consists of three innovative designs. First, it realizes "semantic coherent" synthesis by embedding texts at semantically sensible regions within the background image, where the semantic coherence is achieved by leveraging the semantic annotations of objects and image regions that have been created in the prior semantic segmentation research. Second, it exploits visual saliency to determine the embedding locations within each semantic sensible region, which coincides with the fact that texts are often placed around homogeneous regions for better visibility in scenes. Third, it designs an adaptive text appearance model that determines the color and brightness of embedded texts by learning from the feature of real scene text images adaptively. The proposed technique has been evaluated over five public datasets and the experiments show its superior performance in training accurate and robust scene text detection and recognition models.