We study the settings for which deep contextual embeddings (e.g., BERT) give large improvements in performance relative to classic pretrained embeddings (e.g., GloVe), and an even simpler baseline-random word embeddings-focusing on the impact of the training set size and the linguistic properties of the task.Surprisingly, we find that both of these simpler baselines can match contextual embeddings on industry-scale data, and often perform within 5 to 10% accuracy (absolute) on benchmark tasks.Furthermore, we identify properties of data for which contextual embeddings give particularly large gains: language containing complex structure, ambiguous word usage, and words unseen in training.

We investigate how to train kernel approximation methods that generalize well under a memory budget. Building on recent theoretical work, we define a measure of kernel approximation error which we find to be more predictive of the empirical generalization performance of kernel approximation methods than conventional metrics. An important consequence of this definition is that a kernel approximation matrix must be high rank to attain close approximation. Because storing a high-rank approximation is memory intensive, we propose using a low-precision quantization of random Fourier features (LP-RFFs) to build a high-rank approximation under a memory budget. Theoretically, we show quantization has a negligible effect on generalization performance in important settings. Empirically, we demonstrate across four benchmark datasets that LP-RFFs can match the performance of full-precision RFFs and the Nyström method, with 3x-10x and 50x-460x less memory, respectively.

Compressing word embeddings is important for deploying NLP models in memory-constrained settings. However, understanding what makes compressed embeddings perform well on downstream tasks is challenging---existing measures of compression quality often fail to distinguish between embeddings that perform well and those that do not. We thus propose the eigenspace overlap score as a new measure. We relate the eigenspace overlap score to downstream performance by developing generalization bounds for the compressed embeddings in terms of this score, in the context of linear and logistic regression. We then show that we can lower bound the eigenspace overlap score for a simple uniform quantization compression method, helping to explain the strong empirical performance of this method. Finally, we show that by using the eigenspace overlap score as a selection criterion between embeddings drawn from a representative set we compressed, we can efficiently identify the better performing embedding with up to 2x lower selection error rates than the next best measure of compression quality, and avoid the cost of training a separate model for each task of interest.

This paper presents the first, 15-PetaFLOP Deep Learning system for solving scientific pattern classification problems on contemporary HPC architectures. We develop supervised convolutional architectures for discriminating signals in high-energy physics data as well as semi-supervised architectures for localizing and classifying extreme weather in climate data. Our Intelcaffe-based implementation obtains ~2TFLOP/s on a single Cori Phase-II Xeon-Phi node. We use a hybrid strategy employing synchronous node-groups, while using asynchronous communication across groups. We use this strategy to scale training of a single model to ~9600 Xeon-Phi nodes; obtaining peak performance of 11.73-15.07 PFLOP/s and sustained performance of 11.41-13.27 PFLOP/s. At scale, our HEP architecture produces state-of-the-art classification accuracy on a dataset with 10M images, exceeding that achieved by selections on high-level physics-motivated features. Our semi-supervised architecture successfully extracts weather patterns in a 15TB climate dataset. Our results demonstrate that Deep Learning can be optimized and scaled effectively on many-core, HPC systems.

We present the Stanford Question Answering Dataset (SQuAD), a new reading comprehension dataset consisting of 100,000+ questions posed by crowdworkers on a set of Wikipedia articles, where the answer to each question is a segment of text from the corresponding reading passage. We analyze the dataset to understand the types of reasoning required to answer the questions, leaning heavily on dependency and constituency trees. We build a strong logistic regression model, which achieves an F1 score of 51.0%, a significant improvement over a simple baseline (20%). However, human performance (86.8%) is much higher, indicating that the dataset presents a good challenge problem for future research. The dataset is freely available at https://stanford-qa.com

Higher-order models have been shown to be very useful for a plethora of computer vision tasks. However, existing techniques have focused mainly on MAP inference. In this paper, we present the first efficient approach towards approximate Bayesian marginal inference in a general class of high-order, multi-label attractive models, where previous techniques slow down exponentially with the order (clique size). We formalize this task as performing inference in log-supermodular models under partition constraints, and present an efficient variational inference technique. The resulting optimization problems are convex and yield bounds on the partition function. We also obtain a fully factorized approximation to the posterior, which can be used in lieu of the true complicated distribution. We empirically demonstrate the performance of our approach by comparing it to traditional inference methods on a challenging high-fidelity multi-label image segmentation dataset. We obtain state-of-the-art classification accuracy for MAP inference, and substantially improved ROC curves using the approximate marginals.

In this paper we propose an approach to jointly estimate the layout of rooms as well as the clutter present in the scene using RGB-D data. Towards this goal, we propose an effective model that is able to exploit both depth and appearance features, which are complementary. Furthermore, our approach is efficient as we exploit the inherent decomposition of additive potentials. We demonstrate the effectiveness of our approach on the challenging NYU v2 dataset and show that employing depth reduces the layout error by 6% and the clutter estimation by 13%.