Language model pretraining has led to significant performance gains but careful comparison between different approaches is challenging. Training is computationally expensive, often done on private datasets of different sizes, and, as we will show, hyperparameter choices have significant impact on the final results. We present a replication study of BERT pretraining (Devlin et al., 2019) that carefully measures the impact of many key hyperparameters and training data size. We find that BERT was significantly undertrained, and can match or exceed the performance of every model published after it. Our best model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results highlight the importance of previously overlooked design choices, and raise questions about the source of recently reported improvements. We release our models and code.

/pdf/roberta-a-robustly-optimized-bert-pretraining-approach-2rj0bdyhim.pdf

RoBERTa: A Robustly Optimized BERT Pretraining Approach

While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train.

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Proteins are essential to life, and understanding their structure can facilitate a mechanistic understanding of their function. Through an enormous experimental effort1–4, the structures of around 100,000 unique proteins have been determined5, but this represents a small fraction of the billions of known protein sequences6,7. Structural coverage is bottlenecked by the months to years of painstaking effort required to determine a single protein structure. Accurate computational approaches are needed to address this gap and to enable large-scale structural bioinformatics. Predicting the three-dimensional structure that a protein will adopt based solely on its amino acid sequence—the structure prediction component of the ‘protein folding problem’8—has been an important open research problem for more than 50 years9. Despite recent progress10–14, existing methods fall far short of atomic accuracy, especially when no homologous structure is available. Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known. We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14)15, demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods. Underpinning the latest version of AlphaFold is a novel machine learning approach that incorporates physical and biological knowledge about protein structure, leveraging multi-sequence alignments, into the design of the deep learning algorithm. AlphaFold predicts protein structures with an accuracy competitive with experimental structures in the majority of cases using a novel deep learning architecture.

/pdf/highly-accurate-protein-structure-prediction-with-alphafold-2xjbzfwu3n.pdf

Highly accurate protein structure prediction with AlphaFold

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

Information extraction from microblog posts is an important task, as today microblogs capture an unprecedented amount of information and provide a view into the pulse of the world. As the core component of information extraction, we consider the task of Twitter entity linking in this paper. In the current entity linking literature, mention detection and entity disambiguation are frequently cast as equally important but distinct problems. However, in our task, we find that mention detection is often the performance bottleneck. The reason is that messages on micro-blogs are short, noisy and informal texts with little context, and often contain phrases with ambiguous meanings. To rigorously address the Twitter entity linking problem, we propose a structural SVM algorithm for entity linking that jointly optimizes mention detection and entity disambiguation as a single end-to-end task. By combining structural learning and a variety of firstorder, second-order, and context-sensitive features, our system is able to outperform existing state-of-the art entity linking systems by 15%F1.

/pdf/to-link-or-not-to-link-a-study-on-end-to-end-tweet-entity-3u8ume92wm.pdf

To Link or Not to Link? A Study on End-to-End Tweet Entity Linking

Recent developments in NLP have been accompanied by large, expensive models. Knowledge distillation is the standard method to realize these gains in applications with limited resources: a compact student is trained to recover the outputs of a powerful teacher. While most prior work investigates student architectures and transfer techniques, we focus on an often-neglected aspect---student initialization. We argue that a random starting point hinders students from fully leveraging the teacher expertise, even in the presence of a large transfer set. We observe that applying language model pre-training to students unlocks their generalization potential, surprisingly even for very compact networks. We conduct experiments on 4 NLP tasks and 24 sizes of Transformer-based students; for sentiment classification on the Amazon Book Reviews dataset, pre-training boosts size reduction and TPU speed-up from 3.1x/1.25x to 31x/16x. Extensive ablation studies dissect the interaction between pre-training and distillation, revealing a compound effect even when they are applied on the same unlabeled dataset.

Well-Read Students Learn Better: The Impact of Student Initialization on Knowledge Distillation

Recent work on open domain question answering (QA) assumes strong supervision of the supporting evidence and/or assumes a blackbox information retrieval (IR) system to retrieve evidence candidates. We argue that both are suboptimal, since gold evidence is not always available, and QA is fundamentally different from IR. We show for the first time that it is possible to jointly learn the retriever and reader from question-answer string pairs and without any IR system. In this setting, evidence retrieval from all of Wikipedia is treated as a latent variable. Since this is impractical to learn from scratch, we pre-train the retriever with an Inverse Cloze Task. We evaluate on open versions of five QA datasets. On datasets where the questioner already knows the answer, a traditional IR system such as BM25 is sufficient. On datasets where a user is genuinely seeking an answer, we show that learned retrieval is crucial, outperforming BM25 by up to 19 points in exact match.

Latent Retrieval for Weakly Supervised Open Domain Question Answering

Making complex decisions in real world problems often involves assigning values to sets of interdependent variables where an expressive dependency structure among these can influence, or even dictate, what assignments are possible. Commonly used models typically ignore expressive dependencies since the traditional way of incorporating non-local dependencies is inefficient and hence leads to expensive training and inference.

The contribution of this paper is two-fold. First, this paper presents Constrained Conditional Models (CCMs), a framework that augments linear models with declarative constraints as a way to support decisions in an expressive output space while maintaining modularity and tractability of training. The paper develops, analyzes and compares novel algorithms for CCMs based on Hidden Markov Models and Structured Perceptron. The proposed CCM framework is also compared to task-tailored models, such as semi-CRFs.

Second, we propose CoDL, a constraint-driven learning algorithm, which makes use of constraints to guide semi-supervised learning. We provide theoretical justification for CoDL along with empirical results which show the advantage of using declarative constraints in the context of semi-supervised training of probabilistic models.

/pdf/structured-learning-with-constrained-conditional-models-3ekvxsyq6c.pdf

Structured learning with constrained conditional models

Non-linear models recently receive a lot of attention as people are starting to discover the power of statistical and embedding features. However, tree-based models are seldom studied in the context of structured learning despite their recent success on various classification and ranking tasks. In this paper, we propose S-MART, a tree-based structured learning framework based on multiple additive regression trees. S-MART is especially suitable for handling tasks with dense features, and can be used to learn many different structures under various loss functions. We apply S-MART to the task of tweet entity linking — a core component of tweet information extraction, which aims to identify and link name mentions to entities in a knowledge base. A novel inference algorithm is proposed to handle the special structure of the task. The experimental results show that S-MART significantly outperforms state-of-the-art tweet entity linking systems.

Ming-Wei Chang

Papers

To Link or Not to Link? A Study on End-to-End Tweet Entity Linking

Well-Read Students Learn Better: The Impact of Student Initialization on Knowledge Distillation

Latent Retrieval for Weakly Supervised Open Domain Question Answering

Structured learning with constrained conditional models

S-MART: Novel Tree-based Structured Learning Algorithms Applied to Tweet Entity Linking