Abstract This paper studies the use of language models as a source of synthetic unlabeled text for NLP. We formulate a general framework called “generate, annotate, and learn (GAL)” to take advantage of synthetic text within knowledge distillation, self-training, and few-shot learning applications. To generate high-quality task-specific text, we either fine-tune LMs on inputs from the task of interest, or prompt large LMs with few examples. We use the best available classifier to annotate synthetic text with soft pseudo labels for knowledge distillation and self-training, and use LMs to obtain hard labels for few-shot learning. We train new supervised models on the combination of labeled and pseudo-labeled data, which results in significant gains across several applications. We investigate key components of GAL and present theoretical and empirical arguments against the use of class-conditional LMs to generate synthetic labeled text instead of unlabeled text. GAL achieves new state-of-the-art knowledge distillation results for 6-layer transformers on the GLUE leaderboard.

OBJECTIVE: De-identification is a fundamental task in electronic health records to remove protected health information entities. Deep learning models have proven to be promising tools to automate de-identification processes. However, when the target domain (where the model is applied) is different from the source domain (where the model is trained), the model often suffers a significant performance drop, commonly referred to as domain adaptation issue. In de-identification, domain adaptation issues can make the model vulnerable for deployment. In this work, we aim to close the domain gap by leveraging unlabeled data from the target domain. MATERIALS AND METHODS: We introduce a self-training framework to address the domain adaptation issue by leveraging unlabeled data from the target domain. We validate the effectiveness on 4 standard de-identification datasets. In each experiment, we use a pair of datasets: labeled data from the source domain and unlabeled data from the target domain. We compare the proposed self-training framework with supervised learning that directly deploys the model trained on the source domain. RESULTS: In summary, our proposed framework improves the F1-score by 5.38 (on average) when compared with direct deployment. For example, using i2b2-2014 as the training dataset and i2b2-2006 as the test, the proposed framework increases the F1-score from 76.61 to 85.41 (+8.8). The method also increases the F1-score by 10.86 for mimic-radiology and mimic-discharge. CONCLUSION: Our work demonstrates an effective self-training framework to boost the domain adaptation performance for the de-identification task for electronic health records.

In this work, we present an empirical study of generation order for machine translation.Building on recent advances in insertion-based modeling, we first introduce a soft orderreward framework that enables us to train models to follow arbitrary oracle generation policies.We then make use of this framework to explore a large variety of generation orders, including uninformed orders, locationbased orders, frequency-based orders, contentbased orders, and model-based orders.Curiously, we find that for the WMT'14 English → German and WMT'18 English → Chinese translation tasks, order does not have a substantial impact on output quality.Moreover, for English → German, we even discover that unintuitive orderings such as alphabetical and shortest-first can match the performance of a standard Transformer, suggesting that traditional left-to-right generation may not be necessary to achieve high performance.

We present the Insertion Transformer, an iterative, partially autoregressive model for sequence generation based on insertion operations. Unlike typical autoregressive models which rely on a fixed, often left-to-right ordering of the output, our approach accommodates arbitrary orderings by allowing for tokens to be inserted anywhere in the sequence during decoding. This flexibility confers a number of advantages: for instance, not only can our model be trained to follow specific orderings such as left-to-right generation or a binary tree traversal, but it can also be trained to maximize entropy over all valid insertions for robustness. In addition, our model seamlessly accommodates both fully autoregressive generation (one insertion at a time) and partially autoregressive generation (simultaneous insertions at multiple locations). We validate our approach by analyzing its performance on the WMT 2014 English-German machine translation task under various settings for training and decoding. We find that the Insertion Transformer outperforms many prior non-autoregressive approaches to translation at comparable or better levels of parallelism, and successfully recovers the performance of the original Transformer while requiring only logarithmically many iterations during decoding.

We introduce graph normalizing flows: a new, reversible graph neural network model for prediction and generation. On supervised tasks, graph normalizing flows perform similarly to message passing neural networks, but at a significantly reduced memory footprint, allowing them to scale to larger graphs. In the unsupervised case, we combine graph normalizing flows with a novel graph auto-encoder to create a generative model of graph structures. Our model is permutation-invariant, generating entire graphs with a single feed-forward pass, and achieves competitive results with the state-of-the art auto-regressive models, while being better suited to parallel computing architectures.

We introduce Picturebook, a large-scale lookup operation to ground language via 'snapshots' of our physical world accessed through image search.For each word in a vocabulary, we extract the top-k images from Google image search and feed the images through a convolutional network to extract a word embedding.We introduce a multimodal gating function to fuse our Picturebook embeddings with other word representations.We also introduce Inverse Picturebook, a mechanism to map a Picturebook embedding back into words.We experiment and report results across a wide range of tasks: word similarity, natural language inference, semantic relatedness, sentiment/topic classification, image-sentence ranking and machine translation.We also show that gate activations corresponding to Picturebook embeddings are highly correlated to human judgments of concreteness ratings.

Natural language inference has been shown to be an effective supervised task for learning generic sentence embeddings.In order to better understand the components that lead to effective representations, we propose a lightweight version of InferSent (Conneau et al., 2017), called InferLite, that does not use any recurrent layers and operates on a collection of pre-trained word embeddings.We show that a simple instance of our model that makes no use of context, word ordering or position can still obtain competitive performance on the majority of downstream prediction tasks, with most performance gaps being filled by adding local contextual information through temporal convolutions.Our models can be trained in under 1 hour on a single GPU and allows for fast inference of new representations.Finally we describe a semantic hashing layer that allows our model to learn generic binary codes for sentences.

Books are a rich source of both fine-grained information, how a character, an object or a scene looks like, as well as high-level semantics, what someone is thinking, feeling and how these states evolve through a story. This paper aims to align books to their movie releases in order to provide rich descriptive explanations for visual content that go semantically far beyond the captions available in the current datasets. To align movies and books we propose a neural sentence embedding that is trained in an unsupervised way from a large corpus of books, as well as a video-text neural embedding for computing similarities between movie clips and sentences in the book. We propose a context-aware CNN to combine information from multiple sources. We demonstrate good quantitative performance for movie/book alignment and show several qualitative examples that showcase the diversity of tasks our model can be used for.

Bayesian optimization is an effective methodology for the global optimization of functions with expensive evaluations. It relies on querying a distribution over functions defined by a relatively cheap surrogate model. An accurate model for this distribution over functions is critical to the effectiveness of the approach, and is typically fit using Gaussian processes (GPs). However, since GPs scale cubically with the number of observations, it has been challenging to handle objectives whose optimization requires many evaluations, and as such, massively parallelizing the optimization. In this work, we explore the use of neural networks as an alternative to GPs to model distributions over functions. We show that performing adaptive basis function regression with a neural network as the parametric form performs competitively with state-of-the-art GP-based approaches, but scales linearly with the number of data rather than cubically. This allows us to achieve a previously intractable degree of parallelism, which we apply to large scale hyperparameter optimization, rapidly finding competitive models on benchmark object recognition tasks using convolutional networks, and image caption generation using neural language models.

Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.

We describe an approach for unsupervised learning of a generic, distributed sentence encoder. Using the continuity of text from books, we train an encoder-decoder model that tries to reconstruct the surrounding sentences of an encoded passage. Sentences that share semantic and syntactic properties are thus mapped to similar vector representations. We next introduce a simple vocabulary expansion method to encode words that were not seen as part of training, allowing us to expand our vocabulary to a million words. After training our model, we extract and evaluate our vectors with linear models on 8 tasks: semantic relatedness, paraphrase detection, image-sentence ranking, question-type classification and 4 benchmark sentiment and subjectivity datasets. The end result is an off-the-shelf encoder that can produce highly generic sentence representations that are robust and perform well in practice. We will make our encoder publicly available.

This work aims to address the problem of image-based question-answering (QA) with new models and datasets. In our work, we propose to use neural networks and visual semantic embeddings, without intermediate stages such as object detection and image segmentation, to predict answers to simple questions about images. Our model performs 1.8 times better than the only published results on an existing image QA dataset. We also present a question generation algorithm that converts image descriptions, which are widely available, into QA form. We used this algorithm to produce an order-of-magnitude larger dataset, with more evenly distributed answers. A suite of baseline results on this new dataset are also presented.

Semi-structured documents are a common type of data containing free text in natural language (unstructured data) as well as additional information about the document, or meta-data, typically following a schema or controlled vocabulary (structured data). Simultaneous analysis of unstructured and structured data enables the discovery of hidden relationships that cannot be identified from either of these sources when analyzed independently of each other. In this work, we present a visual text analytics tool for semi-structured documents (ViTA-SSD), that aims to support the user in the exploration and finding of insightful patterns in a visual and interactive manner in a semi-structured collection of documents. It achieves this goal by presenting to the user a set of coordinated visualizations that allows the linking of the metadata with interactively generated clusters of documents in such a way that relevant patterns can be easily spotted. The system contains two novel approaches in its back end: a feature-learning method to learn a compact representation of the corpus and a fast-clustering approach that has been redesigned to allow user supervision. These novel contributions make it possible for the user to interact with a large and dynamic document collection and to perform several text analytical tasks more efficiently. Finally, we present two use cases that illustrate the suitability of the system for in-depth interactive exploration of semi-structured document collections, two user studies, and results of several evaluations of our text-mining components.

We introduce two multimodal neural language models: models of natural language that can be conditioned on other modalities. An image-text multimodal neural language model can be used to retrieve images given complex sentence queries, retrieve phrase descriptions given image queries, as well as generate text conditioned on images. We show that in the case of image-text modelling we can jointly learn word representations and image features by training our models together with a convolutional network. Unlike many of the existing methods, our approach can generate sentence descriptions for images without the use of templates, structured prediction, and/or syntactic trees. While we focus on image-text modelling, our algorithms can be easily applied to other modalities such as audio.

The task of assigning a set of relevant tags to an image is challenging due to the size and variability of tag vocabularies. Consequently, most existing algorithms focus on tag assignment and fix an often large number of hand-crafted features to describe image characteristics. In this paper we introduce a hierarchical model for learning representations of full sized color images from the pixel level, removing the need for engineered feature representations and subsequent feature selection. We benchmark our model on the STL-10 recognition dataset, achieving state-of-the-art performance. When our features are combined with TagProp (Guillaumin et al.), we outperform or compete with existing annotation approaches that use over a dozen distinct image descriptors. Furthermore, using 256-bit codes and Hamming distance for training TagProp, we exchange only a small reduction in performance for efficient storage and fast comparisons. In our experiments, using deeper architectures always outperform shallow ones.