PyTorchKR Papers

최신 논문

총 6,088편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2202.08791공식 코드 · Papers with Code

cosFormer: Rethinking Softmax in Attention

Transformer has shown great successes in natural language processing, computer vision, and audio processing. As one of its core components, the softmax attention helps to capture long-range dependencies yet prohibits its scale-up due to the quadratic space and time complexity to the sequence length. Kernel methods are often adopted to reduce the complexity by approximating the softmax operator. Nevertheless, due to the approximation errors, their performances vary in different tasks/corpus and suffer crucial performance drops when compared with the vanilla softmax attention. In this paper, we propose a linear transformer called cosFormer that can achieve comparable or better accuracy to the vanilla transformer in both casual and cross attentions. cosFormer is based on two key properties of softmax attention: i). non-negativeness of the attention matrix; ii). a non-linear re-weighting scheme that can concentrate the distribution of the attention matrix. As its linear substitute, cosFormer fulfills these properties with a linear operator and a cosine-based distance re-weighting mechanism. Extensive experiments on language modeling and text understanding tasks demonstrate the effectiveness of our method. We further examine our method on long sequences and achieve state-of-the-art performance on the Long-Range Arena benchmark. The source code is available at https://github.com/OpenNLPLab/cosFormer.

arXiv:2202.08370공식 코드 · Papers with Code

CAREER: A Foundation Model for Labor Sequence Data

Labor economists regularly analyze employment data by fitting predictive models to small, carefully constructed longitudinal survey datasets. Although machine learning methods offer promise for such problems, these survey datasets are too small to take advantage of them. In recent years large datasets of online resumes have also become available, providing data about the career trajectories of millions of individuals. However, standard econometric models cannot take advantage of their scale or incorporate them into the analysis of survey data. To this end we develop CAREER, a foundation model for job sequences. CAREER is first fit to large, passively-collected resume data and then fine-tuned to smaller, better-curated datasets for economic inferences. We fit CAREER to a dataset of 24 million job sequences from resumes, and adjust it on small longitudinal survey datasets. We find that CAREER forms accurate predictions of job sequences, outperforming econometric baselines on three widely-used economics datasets. We further find that CAREER can be used to form good predictions of other downstream variables. For example, incorporating CAREER into a wage model provides better predictions than the econometric models currently in use.

arXiv:2202.08005공식 코드 · Papers with Code

Should You Mask 15% in Masked Language Modeling?

Masked language models (MLMs) conventionally mask 15% of tokens due to the belief that more masking would leave insufficient context to learn good representations; this masking rate has been widely used, regardless of model sizes or masking strategies. In this work, we revisit this important choice of MLM pre-training. We first establish that 15% is not universally optimal, and larger models should adopt a higher masking rate. Specifically, we find that masking 40% outperforms 15% for BERT-large size models on GLUE and SQuAD. Interestingly, an extremely high masking rate of 80% can still preserve 95% fine-tuning performance and most of the accuracy in linguistic probing, challenging the conventional wisdom about the role of the masking rate. We then examine the interplay between masking rates and masking strategies and find that uniform masking requires a higher masking rate compared to sophisticated masking strategies such as span or PMI masking. Finally, we argue that increasing the masking rate has two distinct effects: it leads to more corruption, which makes the prediction task more difficult; it also enables more predictions, which benefits optimization. Using this framework, we revisit BERT's 80-10-10 corruption strategy. Together, our results contribute to a better understanding of MLM pre-training.

arXiv:2202.07765공식 코드 · Papers with Code

General-purpose, long-context autoregressive modeling with Perceiver AR

Real-world data is high-dimensional: a book, image, or musical performance can easily contain hundreds of thousands of elements even after compression. However, the most commonly used autoregressive models, Transformers, are prohibitively expensive to scale to the number of inputs and layers needed to capture this long-range structure. We develop Perceiver AR, an autoregressive, modality-agnostic architecture which uses cross-attention to map long-range inputs to a small number of latents while also maintaining end-to-end causal masking. Perceiver AR can directly attend to over a hundred thousand tokens, enabling practical long-context density estimation without the need for hand-crafted sparsity patterns or memory mechanisms. When trained on images or music, Perceiver AR generates outputs with clear long-term coherence and structure. Our architecture also obtains state-of-the-art likelihood on long-sequence benchmarks, including 64 x 64 ImageNet images and PG-19 books.

arXiv:2202.06671공식 코드 · Papers with Code

Neighborhood Contrastive Learning for Scientific Document Representations with Citation Embeddings

Learning scientific document representations can be substantially improved through contrastive learning objectives, where the challenge lies in creating positive and negative training samples that encode the desired similarity semantics. Prior work relies on discrete citation relations to generate contrast samples. However, discrete citations enforce a hard cut-off to similarity. This is counter-intuitive to similarity-based learning, and ignores that scientific papers can be very similar despite lacking a direct citation - a core problem of finding related research. Instead, we use controlled nearest neighbor sampling over citation graph embeddings for contrastive learning. This control allows us to learn continuous similarity, to sample hard-to-learn negatives and positives, and also to avoid collisions between negative and positive samples by controlling the sampling margin between them. The resulting method SciNCL outperforms the state-of-the-art on the SciDocs benchmark. Furthermore, we demonstrate that it can train (or tune) models sample-efficiently, and that it can be combined with recent training-efficient methods. Perhaps surprisingly, even training a general-domain language model this way outperforms baselines pretrained in-domain.

arXiv:2202.06689공식 코드 · Papers with Code

CodeFill: Multi-token Code Completion by Jointly Learning from Structure and Naming Sequences

Code completion is an essential feature of IDEs, yet current autocompleters are restricted to either grammar-based or NLP-based single token completions. Both approaches have significant drawbacks: grammar-based autocompletion is restricted in dynamically-typed language environments, whereas NLP-based autocompleters struggle to understand the semantics of the programming language and the developer's code context. In this work, we present CodeFill, a language model for autocompletion that combines learned structure and naming information. Using a parallel Transformer architecture and multi-task learning, CodeFill consumes sequences of source code token names and their equivalent AST token types. Uniquely, CodeFill is trained both for single-token and multi-token (statement) prediction, which enables it to learn long-range dependencies among grammatical and naming elements. We train CodeFill on two datasets, consisting of 29M and 425M lines of code, respectively. To make the evaluation more realistic, we develop a method to automatically infer points in the source code at which completion matters. We compare CodeFill against four baselines and two state-of-the-art models, GPT-C and TravTrans+.CodeFill surpasses all baselines in single token prediction (MRR: 70.9% vs. 66.2% and 67.8%) and outperforms the state of the art for multi-token prediction (ROUGE-L: 63.7% vs. 52.4% and 59.2%, for n=4 tokens). We publicly release our source code and datasets.

arXiv:2202.06675공식 코드 · Papers with Code

Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content?

Large datasets underlying much of current machine learning raise serious issues concerning inappropriate content such as offensive, insulting, threatening, or might otherwise cause anxiety. This calls for increased dataset documentation, e.g., using datasheets. They, among other topics, encourage to reflect on the composition of the datasets. So far, this documentation, however, is done manually and therefore can be tedious and error-prone, especially for large image datasets. Here we ask the arguably "circular" question of whether a machine can help us reflect on inappropriate content, answering Question 16 in Datasheets. To this end, we propose to use the information stored in pre-trained transformer models to assist us in the documentation process. Specifically, prompt-tuning based on a dataset of socio-moral values steers CLIP to identify potentially inappropriate content, therefore reducing human labor. We then document the inappropriate images found using word clouds, based on captions generated using a vision-language model. The documentations of two popular, large-scale computer vision datasets -- ImageNet and OpenImages -- produced this way suggest that machines can indeed help dataset creators to answer Question 16 on inappropriate image content.

arXiv:2202.05798공식 코드 · Papers with Code

The Dual Form of Neural Networks Revisited: Connecting Test Time Predictions to Training Patterns via Spotlights of Attention

Linear layers in neural networks (NNs) trained by gradient descent can be expressed as a key-value memory system which stores all training datapoints and the initial weights, and produces outputs using unnormalised dot attention over the entire training experience. While this has been technically known since the 1960s, no prior work has effectively studied the operations of NNs in such a form, presumably due to prohibitive time and space complexities and impractical model sizes, all of them growing linearly with the number of training patterns which may get very large. However, this dual formulation offers a possibility of directly visualising how an NN makes use of training patterns at test time, by examining the corresponding attention weights. We conduct experiments on small scale supervised image classification tasks in single-task, multi-task, and continual learning settings, as well as language modelling, and discuss potentials and limits of this view for better understanding and interpreting how NNs exploit training patterns. Our code is public.

arXiv:2202.05607공식 코드 · Papers with Code

Online Decision Transformer

Recent work has shown that offline reinforcement learning (RL) can be formulated as a sequence modeling problem (Chen et al., 2021; Janner et al., 2021) and solved via approaches similar to large-scale language modeling. However, any practical instantiation of RL also involves an online component, where policies pretrained on passive offline datasets are finetuned via taskspecific interactions with the environment. We propose Online Decision Transformers (ODT), an RL algorithm based on sequence modeling that blends offline pretraining with online finetuning in a unified framework. Our framework uses sequence-level entropy regularizers in conjunction with autoregressive modeling objectives for sample-efficient exploration and finetuning. Empirically, we show that ODT is competitive with the state-of-the-art in absolute performance on the D4RL benchmark but shows much more significant gains during the finetuning procedure.

arXiv:2202.05209공식 코드 · Papers with Code

Improving Automatic Speech Recognition for Non-Native English with Transfer Learning and Language Model Decoding

ASR systems designed for native English (L1) usually underperform on non-native English (L2). To address this performance gap, \textbf{(i)} we extend our previous work to investigate fine-tuning of a pre-trained wav2vec 2.0 model \cite{baevski2020wav2vec,xu2021self} under a rich set of L1 and L2 training conditions. We further \textbf{(ii)} incorporate language model decoding in the ASR system, along with the fine-tuning method. Quantifying gains acquired from each of these two approaches separately and an error analysis allows us to identify different sources of improvement within our models. We find that while the large self-trained wav2vec 2.0 may be internalizing sufficient decoding knowledge for clean L1 speech \cite{xu2021self}, this does not hold for L2 speech and accounts for the utility of employing language model decoding on L2 data.

arXiv:2202.04582공식 코드 · Papers with Code

Topic Discovery via Latent Space Clustering of Pretrained Language Model Representations

Topic models have been the prominent tools for automatic topic discovery from text corpora. Despite their effectiveness, topic models suffer from several limitations including the inability of modeling word ordering information in documents, the difficulty of incorporating external linguistic knowledge, and the lack of both accurate and efficient inference methods for approximating the intractable posterior. Recently, pretrained language models (PLMs) have brought astonishing performance improvements to a wide variety of tasks due to their superior representations of text. Interestingly, there have not been standard approaches to deploy PLMs for topic discovery as better alternatives to topic models. In this paper, we begin by analyzing the challenges of using PLM representations for topic discovery, and then propose a joint latent space learning and clustering framework built upon PLM embeddings. In the latent space, topic-word and document-topic distributions are jointly modeled so that the discovered topics can be interpreted by coherent and distinctive terms and meanwhile serve as meaningful summaries of the documents. Our model effectively leverages the strong representation power and superb linguistic features brought by PLMs for topic discovery, and is conceptually simpler than topic models. On two benchmark datasets in different domains, our model generates significantly more coherent and diverse topics than strong topic models, and offers better topic-wise document representations, based on both automatic and human evaluations.

arXiv:2202.03612공식 코드 · Papers with Code

HistBERT: A Pre-trained Language Model for Diachronic Lexical Semantic Analysis

Contextualized word embeddings have demonstrated state-of-the-art performance in various natural language processing tasks including those that concern historical semantic change. However, language models such as BERT was trained primarily on contemporary corpus data. To investigate whether training on historical corpus data improves diachronic semantic analysis, we present a pre-trained BERT-based language model, HistBERT, trained on the balanced Corpus of Historical American English. We examine the effectiveness of our approach by comparing the performance of the original BERT and that of HistBERT, and we report promising results in word similarity and semantic shift analysis. Our work suggests that the effectiveness of contextual embeddings in diachronic semantic analysis is dependent on the temporal profile of the input text and care should be taken in applying this methodology to study historical semantic change.

arXiv:2202.03543공식 코드 · Papers with Code

Self-Supervised Representation Learning for Speech Using Visual Grounding and Masked Language Modeling

In this paper, we describe our submissions to the ZeroSpeech 2021 Challenge and SUPERB benchmark. Our submissions are based on the recently proposed FaST-VGS model, which is a Transformer-based model that learns to associate raw speech waveforms with semantically related images, all without the use of any transcriptions of the speech. Additionally, we introduce a novel extension of this model, FaST-VGS+, which is learned in a multi-task fashion with a masked language modeling objective in addition to the visual grounding objective. On ZeroSpeech 2021, we show that our models perform competitively on the ABX task, outperform all other concurrent submissions on the Syntactic and Semantic tasks, and nearly match the best system on the Lexical task. On the SUPERB benchmark, we show that our models also achieve strong performance, in some cases even outperforming the popular wav2vec2.0 model.

arXiv:2202.04053공식 코드 · Papers with Code

DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Models

Recently, DALL-E, a multimodal transformer language model, and its variants, including diffusion models, have shown high-quality text-to-image generation capabilities. However, despite the realistic image generation results, there has not been a detailed analysis of how to evaluate such models. In this work, we investigate the visual reasoning capabilities and social biases of different text-to-image models, covering both multimodal transformer language models and diffusion models. First, we measure three visual reasoning skills: object recognition, object counting, and spatial relation understanding. For this, we propose PaintSkills, a compositional diagnostic evaluation dataset that measures these skills. Despite the high-fidelity image generation capability, a large gap exists between the performance of recent models and the upper bound accuracy in object counting and spatial relation understanding skills. Second, we assess the gender and skin tone biases by measuring the gender/skin tone distribution of generated images across various professions and attributes. We demonstrate that recent text-to-image generation models learn specific biases about gender and skin tone from web image-text pairs. We hope our work will help guide future progress in improving text-to-image generation models on visual reasoning skills and learning socially unbiased representations. Code and data: https://github.com/j-min/DallEval

arXiv:2202.04003공식 코드 · Papers with Code

Differentiable N-gram Objective on Abstractive Summarization

ROUGE is a standard automatic evaluation metric based on n-grams for sequence-to-sequence tasks, while cross-entropy loss is an essential objective of neural network language model that optimizes at a unigram level. We present differentiable n-gram objectives, attempting to alleviate the discrepancy between training criterion and evaluating criterion. The objective maximizes the probabilistic weight of matched sub-sequences, and the novelty of our work is the objective weights the matched sub-sequences equally and does not ceil the number of matched sub-sequences by the ground truth count of n-grams in reference sequence. We jointly optimize cross-entropy loss and the proposed objective, providing decent ROUGE score enhancement over abstractive summarization dataset CNN/DM and XSum, outperforming alternative n-gram objectives.

arXiv:2202.03829공식 코드 · Papers with Code

TimeLMs: Diachronic Language Models from Twitter

Despite its importance, the time variable has been largely neglected in the NLP and language model literature. In this paper, we present TimeLMs, a set of language models specialized on diachronic Twitter data. We show that a continual learning strategy contributes to enhancing Twitter-based language models' capacity to deal with future and out-of-distribution tweets, while making them competitive with standardized and more monolithic benchmarks. We also perform a number of qualitative analyses showing how they cope with trends and peaks in activity involving specific named entities or concept drift.

arXiv:2202.03632공식 코드 · Papers with Code

ECRECer: Enzyme Commission Number Recommendation and Benchmarking based on Multiagent Dual-core Learning

Enzyme Commission (EC) numbers, which associate a protein sequence with the biochemical reactions it catalyzes, are essential for the accurate understanding of enzyme functions and cellular metabolism. Many ab-initio computational approaches were proposed to predict EC numbers for given input sequences directly. However, the prediction performance (accuracy, recall, precision), usability, and efficiency of existing methods still have much room to be improved. Here, we report ECRECer, a cloud platform for accurately predicting EC numbers based on novel deep learning techniques. To build ECRECer, we evaluate different protein representation methods and adopt a protein language model for protein sequence embedding. After embedding, we propose a multi-agent hierarchy deep learning-based framework to learn the proposed tasks in a multi-task manner. Specifically, we used an extreme multi-label classifier to perform the EC prediction and employed a greedy strategy to integrate and fine-tune the final model. Comparative analyses against four representative methods demonstrate that ECRECer delivers the highest performance, which improves accuracy and F1 score by 70% and 20% over the state-of-the-the-art, respectively. With ECRECer, we can annotate numerous enzymes in the Swiss-Prot database with incomplete EC numbers to their full fourth level. Take UniPort protein "A0A0U5GJ41" as an example (1.14.-.-), ECRECer annotated it with "1.14.11.38", which supported by further protein structure analysis based on AlphaFold2. Finally, we established a webserver (https://ecrecer.biodesign.ac.cn) and provided an offline bundle to improve usability.

arXiv:2202.03371공식 코드 · Papers with Code

Cedille: A large autoregressive French language model

Scaling up the size and training of autoregressive language models has enabled novel ways of solving Natural Language Processing tasks using zero-shot and few-shot learning. While extreme-scale language models such as GPT-3 offer multilingual capabilities, zero-shot learning for languages other than English remain largely unexplored. Here, we introduce Cedille, a large open source auto-regressive language model, specifically trained for the French language. Our results show that Cedille outperforms existing French language models and is competitive with GPT-3 on a range of French zero-shot benchmarks. Furthermore, we provide an in-depth comparison of the toxicity exhibited by these models, showing that Cedille marks an improvement in language model safety thanks to dataset filtering.

arXiv:2202.03052공식 코드 · Papers with Code

OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework

In this work, we pursue a unified paradigm for multimodal pretraining to break the scaffolds of complex task/modality-specific customization. We propose OFA, a Task-Agnostic and Modality-Agnostic framework that supports Task Comprehensiveness. OFA unifies a diverse set of cross-modal and unimodal tasks, including image generation, visual grounding, image captioning, image classification, language modeling, etc., in a simple sequence-to-sequence learning framework. OFA follows the instruction-based learning in both pretraining and finetuning stages, requiring no extra task-specific layers for downstream tasks. In comparison with the recent state-of-the-art vision & language models that rely on extremely large cross-modal datasets, OFA is pretrained on only 20M publicly available image-text pairs. Despite its simplicity and relatively small-scale training data, OFA achieves new SOTAs in a series of cross-modal tasks while attaining highly competitive performances on uni-modal tasks. Our further analysis indicates that OFA can also effectively transfer to unseen tasks and unseen domains. Our code and models are publicly available at https://github.com/OFA-Sys/OFA.

arXiv:2202.02113공식 코드 · Papers with Code

From Discrimination to Generation: Knowledge Graph Completion with Generative Transformer

Knowledge graph completion aims to address the problem of extending a KG with missing triples. In this paper, we provide an approach GenKGC, which converts knowledge graph completion to sequence-to-sequence generation task with the pre-trained language model. We further introduce relation-guided demonstration and entity-aware hierarchical decoding for better representation learning and fast inference. Experimental results on three datasets show that our approach can obtain better or comparable performance than baselines and achieve faster inference speed compared with previous methods with pre-trained language models. We also release a new large-scale Chinese knowledge graph dataset AliopenKG500 for research purpose. Code and datasets are available in https://github.com/zjunlp/PromptKG/tree/main/GenKGC.