Discover published sets by community

The paper introduces ELMo, a deep contextualized word representation that models both complex characteristics of word use and word sense disambiguation, significantly improving the state of the art across a range of challenging NLP tasks.

This pivotal paper presents a neural machine translation model which integrates alignment information, using a novel attention mechanism to improve translation quality.

The paper augments the Stanford Question Answering Dataset (SQuAD) with unanswerable questions and proposes models that can abstain from answering when presented with a question that is not answerable from the given paragraph.

This paper improves sequence labeling tasks by presenting a hybrid neural network architecture that combines bi-directional LSTM, CNN, and CRF layers to effectively capture context and character-level information.

This early work lays the groundwork for neural language modeling, presenting a framework for learning word representations and a statistical language model based on neural networks.

This paper by OpenAI introduces the GPT model, which is pretrained using a language modeling objective on a large corpus and fine-tuned on a task-specific dataset, significantly improving performance across a range of tasks.

The paper presents ULMFiT, a transfer learning technique that applies unsupervised pretraining of a language model followed by fine-tuning on a target task, yielding strong performance on text classification benchmarks.

This paper presents the General Language Understanding Evaluation (GLUE) benchmark, a collection of diverse natural language understanding tasks with the goal of encouraging model generalization across multiple tasks.

This paper introduces a model that learns a self-attention mechanism for sentence embedding, allowing it to capture different aspects of the sentence and provide interpretable representations.

This paper presents BERT, a method of pre-training language representations that obtain state-of-the-art results on a variety of language understanding tasks.

This seminal work by Mikolov et al. introduces the Word2Vec algorithm which efficiently learns vector representations of words that capture syntactic and semantic word relationships.

This critical paper presents the sequence-to-sequence learning framework using RNNs, a breakthrough that enabled a variety of applications in machine translation and question answering.

This paper proposes the Dynamic Memory Network, a model capable of answering questions based on visual and textual inputs by using episodic memory and an attention mechanism to focus on relevant information.

The paper introduces the Transformer model, which eschews recurrence and convolutions in favor of self-attention mechanisms, and achieves state-of-the-art results in machine translation.