In this article, we describe an automatic differentiation module of PyTorch — a library designed to enable rapid research on machine learning models. It builds upon a few projects, most notably Lua Torch, Chainer, and HIPS Autograd [4], and provides a high performance environment with easy access to automatic differentiation of models executed on different devices (CPU and GPU). To make prototyping easier, PyTorch does not follow the symbolic approach used in many other deep learning frameworks, but focuses on differentiation of purely imperative programs, with a focus on extensibility and low overhead. Note that this preprint is a draft of certain sections from an upcoming paper covering all PyTorch features.

/pdf/automatic-differentiation-in-pytorch-gbuafkf7qo.pdf

Automatic differentiation in PyTorch

Deep learning frameworks have often focused on either usability or speed, but not both. PyTorch is a machine learning library that shows that these two goals are in fact compatible: it provides an imperative and Pythonic programming style that supports code as a model, makes debugging easy and is consistent with other popular scientific computing libraries, while remaining efficient and supporting hardware accelerators such as GPUs. 
In this paper, we detail the principles that drove the implementation of PyTorch and how they are reflected in its architecture. We emphasize that every aspect of PyTorch is a regular Python program under the full control of its user. We also explain how the careful and pragmatic implementation of the key components of its runtime enables them to work together to achieve compelling performance. 
We demonstrate the efficiency of individual subsystems, as well as the overall speed of PyTorch on several common benchmarks.

PyTorch: An Imperative Style, High-Performance Deep Learning Library

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires task-specific fine-tuning datasets of thousands or tens of thousands of examples. By contrast, humans can generally perform a new language task from only a few examples or from simple instructions - something which current NLP systems still largely struggle to do. Here we show that scaling up language models greatly improves task-agnostic, few-shot performance, sometimes even reaching competitiveness with prior state-of-the-art fine-tuning approaches. Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters, 10x more than any previous non-sparse language model, and test its performance in the few-shot setting. For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks, as well as several tasks that require on-the-fly reasoning or domain adaptation, such as unscrambling words, using a novel word in a sentence, or performing 3-digit arithmetic. At the same time, we also identify some datasets where GPT-3's few-shot learning still struggles, as well as some datasets where GPT-3 faces methodological issues related to training on large web corpora. Finally, we find that GPT-3 can generate samples of news articles which human evaluators have difficulty distinguishing from articles written by humans. We discuss broader societal impacts of this finding and of GPT-3 in general.

/pdf/language-models-are-few-shot-learners-2fg8gvia7m.pdf

Language Models are Few-Shot Learners

Deep learning frameworks have often focused on either usability or speed, but not both. PyTorch is a machine learning library that shows that these two goals are in fact compatible: it was designed from first principles to support an imperative and Pythonic programming style that supports code as a model, makes debugging easy and is consistent with other popular scientific computing libraries, while remaining efficient and supporting hardware accelerators such as GPUs. In this paper, we detail the principles that drove the implementation of PyTorch and how they are reflected in its architecture. We emphasize that every aspect of PyTorch is a regular Python program under the full control of its user. We also explain how the careful and pragmatic implementation of the key components of its runtime enables them to work together to achieve compelling performance. We demonstrate the efficiency of individual subsystems, as well as the overall speed of PyTorch on several commonly used benchmarks.

/pdf/pytorch-an-imperative-style-high-performance-deep-learning-59vlz5u10y.pdf

Synthetic word analysis is a potentially important but relatively unexplored problem in Chinese natural language processing. Two issues with the conventional pipeline methods involving word segmentation are (1) the lack of a common segmentation standard and (2) the poor segmentation performance on OOV words. These issues may be circumvented if we adopt the view of character-based parsing, providing both internal structures to synthetic words and global structure to sentences in a seamless fashion. However, the accuracy of synthetic word parsing is not yet satisfactory, due to the lack of research. In view of this, we propose and present experiments on several synthetic word parsers. Additionally, we demonstrate the usefulness of incorporating large unlabelled corpora and a dictionary for this task. Our parsers significantly outperform the baseline (a pipeline method).

/pdf/parsing-chinese-synthetic-words-with-a-character-based-58ja829yim.pdf

Parsing Chinese Synthetic Words with a Character-based Dependency Model

We investigate the hypothesis that word representations ought to incorporate both distributional and relational semantics. To this end, we employ the Alternating Direction Method of Multipliers (ADMM), which flexibly optimizes a distributional objective on raw text and a relational objective on WordNet. Preliminary results on knowledge base completion, analogy tests, and parsing show that word representations trained on both objectives can give improvements in some cases.

Incorporating Both Distributional and Relational Semantics in Word Representations

Adapting machine translation systems in the real world is a difficult problem. In contrast to offline training, users cannot provide the type of fine-grained feedback (such as correct translations) typically used for improving the system. Moreover, different users have different translation needs, and even a single user's needs may change over time. 
In this work we take a different approach, treating the problem of adaptation as one of selection. Instead of adapting a single system, we train many translation systems using different architectures, datasets, and optimization methods. Using bandit learning techniques on simulated user feedback, we learn a policy to choose which system to use for a particular translation task. We show that our approach can (1) quickly adapt to address domain changes in translation tasks, (2) outperform the single best system in mixed-domain translation tasks, and (3) make effective instance-specific decisions when using contextual bandit strategies.

Machine Translation System Selection from Bandit Feedback

Standard neural machine translation (NMT) systems operate primarily on words, ignoring lower-level patterns of morphology. We present a character-aware decoder for NMT that can simultaneously work with both word-level and subword-level sequences which is designed to capture such patterns. We achieve character-awareness by augmenting both the softmax and embedding layers of an attention-based encoder-decoder network with convolutional neural networks that operate on spelling of a word (or subword). While character-aware embeddings have been successfully used in the source-side, we find that mixing character-aware embeddings with standard embeddings is crucial in the target-side. Furthermore, we show that a simple approximate softmax layer can be used for large target-side vocabularies which would otherwise require prohibitively large memory. We experiment on the TED multi-target dataset, translating English into 14 typologically diverse languages. We find that in this low-resource setting, the character-aware decoder provides consistent improvements over word-level and subword-level counterparts with BLEU score gains of up to +3.37.

Character-Aware Decoder for Neural Machine Translation

We observe that the development cross-entropy loss of supervised neural machine translation models scales like a power law with the amount of training data and the number of non-embedding parameters in the model. We discuss some practical implications of these results, such as predicting BLEU achieved by large scale models and predicting the ROI of labeling data in low-resource language pairs.

/pdf/data-and-parameter-scaling-laws-for-neural-machine-2usahk48j2.pdf

Kevin Duh

Papers

Parsing Chinese Synthetic Words with a Character-based Dependency Model

Incorporating Both Distributional and Relational Semantics in Word Representations

Machine Translation System Selection from Bandit Feedback

Character-Aware Decoder for Neural Machine Translation

Data and Parameter Scaling Laws for Neural Machine Translation