Deep networks have enabled significant performance gains across domains, but they often suffer from vanishing/exploding gradients. This is especially true for Transformer architectures where depth beyond 12 layers is difficult to train without large datasets and computational budgets. In general, we find that inefficient signal propagation impedes learning in deep networks. In Transformers, multi-head self-attention is the main cause of this poor signal propagation. To facilitate deep signal propagation, we propose ReZero, a simple change to the architecture that initializes an arbitrary layer as the identity map, using a single additional learned parameter per layer. We apply this technique to language modeling and find that we can easily train ReZero-Transformer networks over a hundred layers. When applied to 12 layer Transformers, ReZero converges 56% faster on enwiki8. ReZero applies beyond Transformers to other residual networks, enabling 1,500% faster convergence for deep fully connected networks and 32% faster convergence for a ResNet-56 trained on CIFAR 10.

Ambiguity is inherently present in many machine learning tasks, but especially for sequential models seldom accounted for, as most only output a single prediction. In this work we propose an extension of the Multiple Hypothesis Prediction (MHP) model to handle ambiguous predictions with sequential data, which is of special importance, as often multiple futures are equally likely. Our approach can be applied to the most common recurrent architectures and can be used with any loss function. Additionally, we introduce a novel metric for ambiguous problems, which is better suited to account for uncertainties and coincides with our intuitive understanding of correctness in the presence of multiple labels. We test our method on several experiments and across diverse tasks dealing with time series data, such as trajectory forecasting and maneuver prediction, achieving promising results.

For several purposes in Natural Language Processing (NLP), such as Information Extraction, Sentiment Analysis or Chatbot, Named Entity Recognition (NER) holds an important role as it helps to determine and categorize entities in text into predefined groups such as the names of persons, locations, quantities, organizations or percentages, etc. In this report, we present our experiments on a neural architecture composed of a Conditional Random Field (CRF) layer stacked on top of a Bi-directional LSTM (BI-LSTM) layer for solving NER tasks. Besides, we also employ a fusion input of embedding vectors (Glove, BERT), which are pre-trained on the huge corpus to boost the generalization capacity of the model. Unfortunately, due to the heavy unbalanced distribution cross-training data, both approaches just attained a bad performance on less training samples classes. To overcome this challenge, we introduce an add-on classification model to split sentences into two different sets: Weak and Strong classes and then designing a couple of Bi-LSTM-CRF models properly to optimize performance on each set. We evaluated our models on the test set and discovered that our method can improve performance for Weak classes significantly by using a very small data set (approximately 0.45\%) compared to the rest classes.

With the increase in population densities and environmental awareness, public transport has become an important aspect of urban life. Consequently, large quantities of transportation data are generated, and mining data from smart card use has become a standardized method to understand the travel habits of passengers. Public transport datasets, however, often may lack data integrity; boarding stop information may be missing due to either imperfect acquirement processes or inadequate reporting. As a result, large quantities of observations and even complete sections of cities might be absent from the smart card database. We have developed a machine (supervised) learning method to impute missing boarding stops based on ordinal classification. In addition, we present a new metric, Pareto Accuracy, to evaluate algorithms where classes have an ordinal nature. Results are based on a case study in the Israeli city of Beer Sheva for one month of data. We show that our proposed method significantly notably outperforms current imputation methods and can improve the accuracy and usefulness of large-scale transportation data.

Multi-way data analysis has become an essential tool for capturing underlying structures in higher-order data sets where standard two-way analysis techniques often fail to discover the hidden correlations between variables in multi-way data. We propose a multi-objective variational autoencoder (MVA) method for smart infrastructure damage detection and diagnosis in multi-way sensing data based on the reconstruction probability of autoencoder deep neural network (ADNN). Our method fuses data from multiple sensors in one ADNN at which informative features are being extracted and utilized for damage identification. It generates probabilistic anomaly scores to detect damage, asses its severity and further localize it via a new localization layer introduced in the ADNN. We evaluated our method on multi-way datasets in the area of structural health monitoring for damage diagnosis purposes. The data was collected from our deployed data acquisition system on a cable-stayed bridge in Western Sydney and from a laboratory based building structure obtained from Los Alamos National Laboratory (LANL). Experimental results show that the proposed method can accurately detect structural damage. It was also able to estimate the different levels of damage severity, and capture damage locations in an unsupervised aspect. Compared to the state-of-the-art approaches, our proposed method shows better performance in terms of damage detection and localization.

We look at a dataset of tropical storm data in the Atlantic Ocean from 1982 to 2017 and perform deep learning predictions with uncertainty bounds on trajectories of the storms. The result of these storms, particularly the strongest ones called hurricanes--defined as having wind speeds exceeding 74 mph--can be devastating because of their strong winds and heavy precipitation that can cause dangerous tides. Tropical storms can cause major environmental disasters when they reach land, such as the 2005 Hurricane Katrina that resulted in over 850 deaths and caused major economic damage and the 2012 Hurricane Sandy that caused almost $70 billion in damage across much of the eastern United States, with peak winds of 115 mph (Hurricane). According to the National Oceanic and Atmospheric Administration, it is likely that global warming will cause hurricanes in the upcoming century to be more intense by 1 to 10% globally (with higher peak winds and lower central pressures), which will result in a higher proportion of more severe storms (NOAA). Historically, hurricane trajectory predictions have used statistical methods that can be limiting because of the nonlinearity and complexity of atmospheric systems. Deep learning techniques and specifically recurrent neural networks have grown in popularity in recent years as a strong method for approaching prediction problems because of the ability to extract important features and relationships from complex high-dimensional data, especially for forecasting and classification (McDermott and Wikle, 2019). We implemented a number of improvements over previous deep learning prediction work (Alemany et al., 2019), including predicting exact storm locations in latitude/longitude instead of a grid value and using a prediction window that uses all previous hurricane data rather than a fixed-size sliding window. While hurricane trajectory predictions have seen improvements recently (SHIPS), we build on previous work to include a fundamental uncertainty measure in the prediction for the first time as part of a neural network framework. The uncertainty measure is especially valuable for understanding a defined location range rather than only a point estimate, which is important for evacuation and safety/preparation purposes.

Accurate models of the world are built upon notions of its underlying symmetries. In physics, these symmetries correspond to conservation laws, such as for energy and momentum. Yet even though neural network models see increasing use in the physical sciences, they struggle to learn these symmetries. In this paper, we propose Lagrangian Neural Networks (LNNs), which can parameterize arbitrary Lagrangians using neural networks. In contrast to models that learn Hamiltonians, LNNs do not require canonical coordinates, and thus perform well in situations where canonical momenta are unknown or difficult to compute. Unlike previous approaches, our method does not restrict the functional form of learned energies and will produce energy-conserving models for a variety of tasks. We test our approach on a double pendulum and a relativistic particle, demonstrating energy conservation where a baseline approach incurs dissipation and modeling relativity without canonical coordinates where a Hamiltonian approach fails. Finally, we show how this model can be applied to graphs and continuous systems using a Lagrangian Graph Network, and demonstrate it on the 1D wave equation.

Precision beekeeping allows to monitor bees' living conditions by equipping beehives with sensors. The data recorded by these hives can be analyzed by machine learning models to learn behavioral patterns of or search for unusual events in bee colonies. One typical target is the early detection of bee swarming as apiarists want to avoid this due to economical reasons. Advanced methods should be able to detect any other unusual or abnormal behavior arising from illness of bees or from technical reasons, e.g. sensor failure. In this position paper we present an autoencoder, a deep learning model, which detects any type of anomaly in data independent of its origin. Our model is able to reveal the same swarms as a simple rule-based swarm detection algorithm but is also triggered by any other anomaly. We evaluated our model on real world data sets that were collected on different hives and with different sensor setups.

Channel attention mechanisms, as the key components of some modern convolutional neural networks (CNNs) architectures, have been commonly used in many visual tasks for effective performance improvement. It is able to reinforce the informative channels and to suppress useless channels of feature maps obtained by CNNs. Recently, different attention modules have been proposed, which are implemented in various ways. However, they are mainly based on convolution and pooling operations, which are lack of intuitive and reasonable insights about the principles that they are based on. Moreover, the ways that they improve the performance of the CNNs is not clear either. In this paper, we propose a Gaussian process embedded channel attention (GPCA) module and interpret the channel attention intuitively and reasonably in a probabilistic way. The GPCA module is able to model the correlations from channels which are assumed as beta distributed variables with Gaussian process prior. As the beta distribution is intractably integrated into the end-to-end training of the CNNs, we utilize an appropriate approximation of the beta distribution to make the distribution assumption implemented easily. In this case, the proposed GPCA module can be integrated into the end-to-end training of the CNNs. Experimental results demonstrate that the proposed GPCA module can improve the accuracies of image classification on four widely used datasets.

Deep learning models on graphs have achieved remarkable performance in various graph analysis tasks, e.g., node classification, link prediction and graph clustering. However, they expose uncertainty and unreliability against the well-designed inputs, i.e., adversarial examples. Accordingly, various studies have emerged for both attack and defense addressed in different graph analysis tasks, leading to the arms race in graph adversarial learning. For instance, the attacker has poisoning and evasion attack, and the defense group correspondingly has preprocessing- and adversarial- based methods. Despite the booming works, there still lacks a unified problem definition and a comprehensive review. To bridge this gap, we investigate and summarize the existing works on graph adversarial learning tasks systemically. Specifically, we survey and unify the existing works w.r.t. attack and defense in graph analysis tasks, and give proper definitions and taxonomies at the same time. Besides, we emphasize the importance of related evaluation metrics, and investigate and summarize them comprehensively. Hopefully, our works can serve as a reference for the relevant researchers, thus providing assistance for their studies. More details of our works are available at https://github.com/gitgiter/Graph-Adversarial-Learning.