Graph neural networks (GNNs) have been widely used in representation learning on graphs and achieved state-of-the-art performance in tasks such as node classification and link prediction. However, most existing GNNs are designed to learn node representations on the fixed and homogeneous graphs. The limitations especially become problematic when learning representations on a misspecified graph or a heterogeneous graph that consists of various types of nodes and edges. In this paper, we propose Graph Transformer Networks (GTNs) that are capable of generating new graph structures, which involve identifying useful connections between unconnected nodes on the original graph, while learning effective node representation on the new graphs in an end-to-end fashion. Graph Transformer layer, a core layer of GTNs, learns a soft selection of edge types and composite relations for generating useful multi-hop connections so-called meta-paths. Our experiments show that GTNs learn new graph structures, based on data and tasks without domain knowledge, and yield powerful node representation via convolution on the new graphs. Without domain-specific graph preprocessing, GTNs achieved the best performance in all three benchmark node classification tasks against the state-of-the-art methods that require pre-defined meta-paths from domain knowledge.

Yingce Xia, Xu T an, Fei Tian, Fei Gao, Weicong Chen, Y ang Fan, Linyuan Gong, Yichong Leng, Renqian Luo, Yiren Wang, Lijun Wu, Jinhua Zhu, T ao Qin, Tie-Y an Liu Microsoft Research Asia Abstract We Microsoft Research Asia made submissions to 11 language directions in the WMT19 news translation tasks. We won the first place for 8 of the 11 directions and the second place for the other three. Our basic systems are built on Transformer, back translation and knowledge distillation. We integrate several of our rececent techniques to enhance the baseline systems: multi-agent dual learning (MADL), masked sequence-to-sequence pre-training (MASS), neural architecture optimization (NAO), and soft contextual data augmentation (SCA). 1 Introduction We participated in the WMT19 shared news translation task in 11 translation directions. We achieved first place for 8 directions: German English, German French, Chinese English, English Lithuanian, English Finnish, and Russian English, and three other directions were placed second (ranked by teams), which included Lithuanian English, Finnish English, and English Kazakh. Our basic systems are based on Transformer, back translation and knowledge distillation. We experimented with several techniques we proposed recently. In brief, the innovations we introduced are: Multi-agent dual learning (MADL) The core idea of dual learning is to leverage the duality between the primal task (mapping from domain X to domain Y) and dual task (mapping from domain Y to X) to boost the performances of both tasks. MADL (Wang et al., 2019) extends the dual learning (He et al., 2016; Xia et al., 2017a) framework by introducing multiple primal and dual models. It was integrated into our submitted systems for*Corresponding author.

The dominant automatic lexical stress detection method is to split the utterance into syllable segments using phoneme sequence and their time-aligned boundaries. Then we extract features from syllable to use classification method to classify the lexical stress. However, we can't get very accurate time boundaries of each phoneme and we have to design some features in the syllable segments to classify the lexical stress. Therefore, we propose a end-to-end approach using sequence to sequence model of transformer to estimate lexical stress. For this, we train transformer model using feature sequence of audio and their phoneme sequence with lexical stress marks. During the recognition process, the recognized phoneme sequence is restricted according to the original standard phoneme sequence without lexical stress marks, but the lexical stress mark of each phoneme is not limited. We train the model in different subset of Librispeech and do lexical stress recognition in TIMIT and L2-ARCTIC dataset. For all subsets, the end-to-end model will perform better than the syllable segments classification method. Our method can achieve a 6.36% phoneme error rate on the TIMIT dataset, which exceeds the 7.2% error rate in other studies.

We examine the effects of instantiating Lewis signaling games within a population of speaker and listener agents with the aim of producing a set of general and robust representations of unstructured pixel data. Preliminary experiments suggest that the set of representations associated with languages generated within a population outperform those generated between a single speaker-listener pair on this objective, making a case for the adoption of population-based approaches in emergent communication studies. Furthermore, post-hoc analysis reveals that population-based learning induces a number of novel factors to the conventional emergent communication setup, inviting a wide range of future research questions regarding communication dynamics and the flow of information within them.

Data sets for fairness relevant tasks can lack examples or be biased according to a specific label in a sensitive attribute. We demonstrate the usefulness of weight based meta-learning approaches in such situations. For models that can be trained through gradient descent, we demonstrate that there are some parameter configurations that allow models to be optimized from a few number of gradient steps and with minimal data which are both fair and accurate. To learn such weight sets, we adapt the popular MAML algorithm to Fair-MAML by the inclusion of a fairness regularization term. In practice, Fair-MAML allows practitioners to train fair machine learning models from only a few examples when data from related tasks is available. We empirically exhibit the value of this technique by comparing to relevant baselines.

This article presents a novel method for predicting suicidal ideation from Electronic Health Records (EHR) and Ecological Momentary Assessment (EMA) data using deep sequential models. Both EHR longitudinal data and EMA question forms are defined by asynchronous, variable length, randomly-sampled data sequences. In our method, we model each of them with a Recurrent Neural Network (RNN), and both sequences are aligned by concatenating the hidden state of each of them using temporal marks. Furthermore, we incorporate attention schemes to improve performance in long sequences and time-independent pre-trained schemes to cope with very short sequences. Using a database of 1023 patients, our experimental results show that the addition of EMA records boosts the system recall to predict the suicidal ideation diagnosis from 48.13% obtained exclusively from EHR-based state-of-the-art methods to 67.78%. Additionally, our method provides interpretability through the t-SNE representation of the latent space. Further, the most relevant input features are identified and interpreted medically.

The quality of neural machine translation can be improved by leveraging additional monolingual resources to create synthetic training data. Source-side monolingual data can be (forward-)translated into the target language for self-training; target-side monolingual data can be back-translated. It has been widely reported that back-translation delivers superior results, but could this be due to artefacts in the test sets? W e perform a case study using French-English news translation task and separate test sets based on their original languages. W e show that forward translation delivers superior gains in terms of BLEU on sentences that were originally in the source language, complementing previous studies which show large improvements with back-translation on sentences that were originally in the target language. To better understand when and why forward and back-translation are effective, we study the role of domains, translationese, and noise. While translationese effects are well known to influence MT evaluation, we also find evidence that news data from different languages shows subtle domain differences, which is another explanation for varying performance on different portions of the test set. W e perform additional low-resource experiments which demonstrate that forward translation is more sensitive to the quality of the initial translation system than back-translation, and tends to perform worse in low-resource settings.

Accelerated design, optimization, and tuning of materials via machine learning is receiving increasing interest in science and industry. A major driver of this interest is the potential to reduce the substantial cost and effort involved in manual development, synthesis, and characterization of large numbers of candidate materials. The primary aim is to reduce the number of both failed candidates and development cycles. A data-driven approach to achieve this acceleration is active learning (AL) [23], an iterative procedure in which a machine-learning model suggests candidate materials, a selection of which are synthesized, characterized, and fed back into the model to complete a learning iteration. The objective of this procedure varies; in materials informatics it is often to identify promising material candidates by optimizing properties of interest.

Ultrasonic guided waves are commonly used to localize structural damage in infrastructures such as buildings, airplanes, bridges. Damage localization can be viewed as an inverse problem. Physical model based techniques are popular for guided wave based damage localization. The performance of these techniques depend on the degree of faithfulness with which the physical model describes wave propagation. External factors such as environmental variations and random noise are a source of uncertainty in wave propagation. The physical modeling of uncertainty in an inverse problem is still a challenging problem. In this work, we propose a deep learning based model for robust damage localization in presence of uncertainty. Wave data with uncertainty is simulated to reflect variations due to external factors and Gaussian noise is added to reflect random noise in the environment. After evaluating the localization error on test data with uncertainty, we observe that the deep learning model trained with uncertainty can learn robust representations. The approach shows potential for dealing with uncertainty in physical science problems using deep learning models.

Our interest focuses on developing statistical methods for analysis of brain structural connectomes. Nodes in the brain connectome graph correspond to different regions of interest (ROIs) while edges correspond to white matter fiber connections between these ROIs. Due to the high-dimensionality and non-Euclidean nature of the data, it becomes challenging to conduct analyses of the population distribution of brain connectomes and relate connectomes to other factors, such as cognition. Current approaches focus on summarizing the graph using either pre-specified topological features or principal components analysis (PCA). In this article, we instead develop a nonlinear latent factor model for summarizing the brain graph in both unsupervised and supervised settings. The proposed approach builds on methods for hierarchical modeling of replicated graph data, as well as variational auto-encoders that use neural networks for dimensionality reduction. We refer to our method as Graph AuTo-Encoding (GATE). We compare GATE with tensor PCA and other competitors through simulations and applications to data from the Human Connectome Project (HCP).