We introduce an efficient approach for optimization over orthogonal groups on highly parallel computation units such as GPUs or TPUs. As in earlier work, we parametrize an orthogonal matrix as a product of Householder reflections. However, to overcome low parallelization capabilities of computing Householder reflections sequentially, we propose employing an accumulation scheme called the compact WY (or CWY) transform -- a compact parallelization-friendly matrix representation for the series of Householder reflections. We further develop a novel Truncated CWY (or T-CWY) approach for Stiefel manifold parametrization which has a competitive complexity and, again, yields benefits when computed on GPUs and TPUs. We prove that our CWY and T-CWY methods lead to convergence to a stationary point of the training objective when coupled with stochastic gradient descent. We apply our methods to train recurrent neural network architectures in the tasks of neural machine translation and video prediction, and demonstrate superiority compared to earlier methods.

The high energy costs of neural network training and inference led to the use of acceleration hardware such as GPUs and TPUs. While this enabled us to train large-scale neural networks in datacenters and deploy them on edge devices, the focus so far is on average-case performance. In this work, we introduce a novel threat vector against neural networks whose energy consumption or decision latency are critical. We show how adversaries can exploit carefully crafted $\boldsymbol{sponge}~\boldsymbol{examples}$, which are inputs designed to maximise energy consumption and latency. We mount two variants of this attack on established vision and language models, increasing energy consumption by a factor of 10 to 200. Our attacks can also be used to delay decisions where a network has critical real-time performance, such as in perception for autonomous vehicles. We demonstrate the portability of our malicious inputs across CPUs and a variety of hardware accelerator chips including GPUs, and an ASIC simulator. We conclude by proposing a defense strategy which mitigates our attack by shifting the analysis of energy consumption in hardware from an average-case to a worst-case perspective.

Classification algorithms in machine learning often assume a flat label space. However, most real world data have dependencies between the labels, which can often be captured by using a hierarchy. Utilizing this relation can help develop a model capable of satisfying the dependencies and improving model accuracy and interpretability. Further, as different levels in the hierarchy correspond to different granularities, penalizing each label equally can be detrimental to model learning. In this paper, we propose a loss function, hierarchical curriculum loss, with two properties: (i) satisfy hierarchical constraints present in the label space, and (ii) provide non-uniform weights to labels based on their levels in the hierarchy, learned implicitly by the training paradigm. We theoretically show that the proposed loss function is a tighter bound of 0-1 loss compared to any other loss satisfying the hierarchical constraints. We test our loss function on real world image data sets, and show that it significantly substantially outperforms multiple baselines.

Predicting faults before they occur helps to avoid potential safety hazards. Furthermore, planning the required maintenance actions in advance reduces operation costs. In this article, the focus is on electrochemical cells. In order to predict a cell's fault, the typical approach is to estimate the expected voltage that a healthy cell would present and compare it with the cell's measured voltage in real-time. This approach is possible because, when a fault is about to happen, the cell's measured voltage differs from the one expected for the same operating conditions. However, estimating the expected voltage is challenging, as the voltage of a healthy cell is also affected by its degradation -- an unknown parameter. Expert-defined parametric models are currently used for this estimation task. Instead, we propose the use of a neural network model based on an encoder-decoder architecture. The network receives the operating conditions as input. The encoder's task is to find a faithful representation of the cell's degradation and to pass it to the decoder, which in turn predicts the expected cell's voltage. As no labeled degradation data is given to the network, we consider our approach to be a self-supervised encoder. Results show that we were able to predict the voltage of multiple cells while diminishing the prediction error that was obtained by the parametric models by 53%. This improvement enabled our network to predict a fault 31 hours before it happened, a 64% increase in reaction time compared to the parametric model. Moreover, the output of the encoder can be plotted, adding interpretability to the neural network model.

Predicting the structure of multi-protein complexes is a grand challenge in biochemistry, with major implications for basic science and drug discovery. Computational structure prediction methods generally leverage pre-defined structural features to distinguish accurate structural models from less accurate ones. This raises the question of whether it is possible to learn characteristics of accurate models directly from atomic coordinates of protein complexes, with no prior assumptions. Here we introduce a machine learning method that learns directly from the 3D positions of all atoms to identify accurate models of protein complexes, without using any pre-computed physics-inspired or statistical terms. Our neural network architecture combines multiple ingredients that together enable end-to-end learning from molecular structures containing tens of thousands of atoms: a point-based representation of atoms, equivariance with respect to rotation and translation, local convolutions, and hierarchical subsampling operations. When used in combination with previously developed scoring functions, our network substantially improves the identification of accurate structural models among a large set of possible models. Our network can also be used to predict the accuracy of a given structural model in absolute terms. The architecture we present is readily applicable to other tasks involving learning on 3D structures of large atomic systems.

Several custom web scrapers were created for retrieving news articles from various online news organizations. All web scrapers were run every two hours to retrieve articles from the following five news sites: the Atlantic, the British Broadcasting Corporation (BBC) News, Fox News, the New York Times and Slate Magazine. Web scrapers continue to run every two hours in perpetuity, scraping additional news articles. Collectively, the web scrapers used each news organization's RSS feed as input, storing the scraped output into a custom database. Article URLs were used for disambiguation; where two scraped articles shared a URL, the most recently retrieved article replaced previous versions of articles. As of November 2019, we collected a total of 105,000 news articles from five media organizations. Figure 2 depicts the number of cumulative articles scraped for each news organization over time. Even though articles from Fox News were regularly scraped four months later than other news sources, the number of articles scraped rose quickly, and now constitutes the news organization with the most scraped articles. Given the news scrapers run at regularly scheduled two-hour intervals for all news organization, this suggests that Fox News updates its RSS feed with new articles far more often than others, and the Atlantic updates its RSS feed far less frequently than others.

Deep clustering algorithms combine representation learning and clustering by jointly optimizing a clustering loss and a non-clustering loss. In such methods, a deep neural network is used for representation learning together with a clustering network. Instead of following this framework to improve clustering performance, we propose a simpler approach of optimizing the entanglement of the learned latent code representation of an autoencoder. We define entanglement as how close pairs of points from the same class or structure are, relative to pairs of points from different classes or structures. To measure the entanglement of data points, we use the soft nearest neighbor loss, and expand it by introducing an annealing temperature factor. Using our proposed approach, the test clustering accuracy was 96.2% on the MNIST dataset, 85.6% on the Fashion-MNIST dataset, and 79.2% on the EMNIST Balanced dataset, outperforming our baseline models.

Prognostics is concerned with predicting the future health of the equipment and any potential failures. With the advances in the Internet of Things (IoT), data-driven approaches for prognostics that leverage the power of machine learning models are gaining popularity. One of the most important categories of data-driven approaches relies on a predefined or learned health indicator to characterize the equipment condition up to the present time and make inference on how it is likely to evolve in the future. In these approaches, health indicator forecasting that constructs the health indicator curve over the lifespan using partially observed measurements (i.e., health indicator values within an initial period) plays a key role. Existing health indicator forecasting algorithms, such as the functional Empirical Bayesian approach, the regression-based formulation, a naive scenario matching based on the nearest neighbor, have certain limitations. In this paper, we propose a new `generative + scenario matching' algorithm for health indicator forecasting. The key idea behind the proposed approach is to first non-parametrically fit the underlying health indicator curve with a continuous Gaussian Process using a sample of run-to-failure health indicator curves. The proposed approach then generates a rich set of random curves from the learned distribution, attempting to obtain all possible variations of the target health condition evolution process over the system's lifespan. The health indicator extrapolation for a piece of functioning equipment is inferred as the generated curve that has the highest matching level within the observed period. Our experimental results show the superiority of our algorithm over the other state-of-the-art methods.

Designing appropriate variational regularization schemes is a crucial part of solving inverse problems, making them better-posed and guaranteeing that the solution of the associated optimization problem satisfies desirable properties. Recently, learning-based strategies have appeared to be very efficient for solving inverse problems, by learning direct inversion schemes or plug-and-play regularizers from available pairs of true states and observations. In this paper, we go a step further and design an end-to-end framework allowing to learn actual variational frameworks for inverse problems in such a supervised setting. The variational cost and the gradient-based solver are both stated as neural networks using automatic differentiation for the latter. We can jointly learn both components to minimize the data reconstruction error on the true states. This leads to a data-driven discovery of variational models. We consider an application to inverse problems with incomplete datasets (image inpainting and multivariate time series interpolation). We experimentally illustrate that this framework can lead to a significant gain in terms of reconstruction performance, including w.r.t. the direct minimization of the variational formulation derived from the known generative model.

Myoelectric control is one of the leading brain-machine-interfaces in the field of robotic prosthetics. We present our research in real-time surface electromyography (sEMG) signal classification, where our simple and novel attention-based approach now leads the industry, universally beating more complex, state-of-the-art models. Our model achieved an accuracy of 87\% (class-balanced accuracy: 69\%) using sEMG data and 91\% (balanced accuracy: 74\%) using both sEMG and accelerometer (IMU) data on NinaPro DB5, as well as 73\% overall on NinaPro DB4, an improvement on both highly sophisticated deep learning and signal processing approaches. Notably, the representation of the data learned by the attention mechanism alone is powerful enough to yield an accuracy of 79\% on DB5. NinaPro DB5 is a standard benchmark for sEMG gesture recognition and consists of 53 unique gestures, including finger gestures, wrist gestures, and functional grasping gestures. Our proposed methodology's model simplicity represents a compelling alternative to the convolutional neural network (CNN) approaches utilized in recent research.