Fine-tuning pre-trained language models (PTLMs), such as BERT and its better variant RoBERTa, has been a common practice for advancing performance in natural language understanding (NLU) tasks. Recent advance in representation learning shows that isotropic (i.e., unit-variance and uncorrelated) embeddings can significantly improve performance on downstream tasks with faster convergence and better generalization. The isotropy of the pre-trained embeddings in PTLMs, however, is relatively under-explored. In this paper, we analyze the isotropy of the pre-trained [CLS] embeddings of PTLMs with straightforward visualization, and point out two major issues: high variance in their standard deviation, and high correlation between different dimensions. We also propose a new network regularization method, isotropic batch normalization (IsoBN) to address the issues, towards learning more isotropic representations in fine-tuning by dynamically penalizing dominating principal components. This simple yet effective fine-tuning method yields about 1.0 absolute increment on the average of seven NLU tasks.
In this work, we study the large-scale pretraining of BERT-Large with differentially private SGD (DP-SGD). We show that combined with a careful implementation, scaling up the batch size to millions (i.e., mega-batches) improves the utility of the DP-SGD step for BERT; we also enhance its efficiency by using an increasing batch size schedule. Our implementation builds on the recent work of [SVK20], who demonstrated that the overhead of a DP-SGD step is minimized with effective use of JAX [BFH+18, FJL18] primitives in conjunction with the XLA compiler [XLA17]. Our implementation achieves a maske…
Google Brain（谷歌大脑）；Google（谷歌）；UC Berkeley（加州大学伯克利分校）；UCLA（加州大学洛杉矶分校）
Training large deep neural networks on massive datasets is computationally very challenging. There has been recent surge in interest in using large batch stochastic optimization methods to tackle this issue. The most prominent algorithm in this line of research is LARS, which by employing layerwise adaptive learning rates trains ResNet on ImageNet in a few minutes. However, LARS performs poorly for attention models like BERT, indicating that its performance gains are not consistent across tasks. In this paper, we first study a principled layerwise adaptation strategy to accelerate training of …
BERT has recently attracted a lot of attention in natural language understanding (NLU) and achieved state-of-the-art results in various NLU tasks. However, its success requires large deep neural networks and huge amount of data, which result in long training time and impede development progress. Using stochastic gradient methods with large mini-batch has been advocated as an efficient tool to reduce the training time. Along this line of research, LAMB is a prominent example that reduces the training time of BERT from 3 days to 76 minutes on a TPUv3 Pod. In this paper, we propose an accelerated…
Recent advances in automatic evaluation metrics for text have shown that deep contextualized word representations, such as those generated by BERT encoders, are helpful for designing metrics that correlate well with human judgements. At the same time, it has been argued that contextualized word representations exhibit sub-optimal statistical properties for encoding the true similarity between words or sentences. In this paper, we present two techniques for improving encoding representations for similarity metrics: a batch-mean centering strategy that improves statistical properties; and a comp…
Large Transformer-based language models are pre-trained on corpora of varying sizes, for a different number of steps and with different batch sizes. At the same time, more fundamental components, such as the pre-training objective or architectural hyperparameters, are modified. In total, it is therefore difficult to ascribe changes in performance to specific factors. Since searching the hyperparameter space over the full systems is too costly, we pre-train down-scaled versions of several popular Transformer-based architectures on a common pre-training corpus and benchmark them on a subset of t…
Effective training of today's large language models (LLMs) depends on large batches and long sequences for throughput and accuracy. To handle variable-length sequences on hardware accelerators, it is common practice to introduce padding tokens, so that all sequences in a batch have the same length. We show in this paper that the variation in sequence lengths in common NLP datasets is such that up to 50% of all tokens can be padding. In less common, but not extreme, cases (e.g. GLUE-cola with sequence length 128), the ratio is up to 89%. Existing methods to address the resulting inefficiency ar…
Self-supervised learning, a.k.a., pretraining, is important in natural language processing. Most of the pretraining methods first randomly mask some positions in a sentence and then train a model to recover the tokens at the masked positions. In such a way, the model can be trained without human labeling, and the massive data can be used with billion parameters. Therefore, the optimization efficiency becomes critical. In this paper, we tackle the problem from the view of gradient variance reduction. In particular, we first propose a principled gradient variance decomposition theorem, which sho…
In recent years, large pre-trained Transformer-based language models have led to dramatic improvements in many natural language understanding tasks. To train these models with increasing sizes, many neural network practitioners attempt to increase the batch sizes in order to leverage multiple GPUs to improve training speed. However, increasing the batch size often makes the optimization more difficult, leading to slow convergence or poor generalization that can require orders of magnitude more training time to achieve the same model quality. In this paper, we explore the steepness of the loss …
The explosion in novel NLP word embedding and deep learning techniques has induced significant endeavors into potential applications. One of these directions is in the financial sector. Although there is a lot of work done in state-of-the-art models like GPT and BERT, there are relatively few works on how well these methods perform through fine-tuning after being pre-trained, as well as info on how sensitive their parameters are. We investigate the performance and sensitivity of transferred neural architectures from pre-trained GPT-2 and BERT models. We test the fine-tuning performance based o…
Domain-Adapted Pre-trained Language Models for Implicit Information Extraction in Crash Narratives
Department of Technology, Management and Economics, Technical University of Denmark（技术、管理与经济系，丹麦技术大学）；Department of Mechanics and Maritime Sciences, Chalmers University of Technology（机械与航海科学系，查尔姆斯理工大学）；Department of Computer Science and Engineering, Chalmers University of Technology（计算机科学与工程系，查尔姆斯理工大学）
Recent advances in pre-training huge models on large amounts of text through self supervision have obtained state-of-the-art results in various natural language processing tasks. However, these huge and expensive models are difficult to use in practise for downstream tasks. Some recent efforts use knowledge distillation to compress these models. However, we see a gap between the performance of the smaller student models as compared to that of the large teacher. In this work, we leverage large amounts of in-domain unlabeled transfer data in addition to a limited amount of labeled training insta…
University of Texas at Austin（德克萨斯大学奥斯汀分校）；Microsoft Corporation（微软公司）
Heavily overparameterized language models such as BERT, XLNet and T5 have achieved impressive success in many NLP tasks. However, their high model complexity requires enormous computation resources and extremely long training time for both pre-training and fine-tuning. Many works have studied model compression on large NLP models, but only focusing on reducing inference time while still requiring an expensive training process. Other works use extremely large batch sizes to shorten the pre-training time, at the expense of higher computational resource demands. In this paper, inspired by the Ear…
Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）；School of Cyber Security, University of Chinese Academy of Sciences（中国科学院大学网络空间安全学院）；Kuaishou Technology（快手科技）
Contrastive learning has been gradually applied to learn high-quality unsupervised sentence embedding. Among the previous un-supervised methods, the latest state-of-the-art method, as far as we know, is unsupervised SimCSE (unsup-SimCSE). Unsup-SimCSE uses the InfoNCE1loss function in the training stage by pulling semantically similar sentences together and pushing apart dis-similar ones.Theoretically, we expect to use larger batches in unsup-SimCSE to get more adequate comparisons among samples and avoid overfitting. However, increasing the batch size does not always lead to improvements, but…
Iran University of Science and Technology（伊朗科技大学）；Islamic Azad University（伊斯兰阿扎德大学）
Multiple instance learning (MIL) has become the standard learning paradigm for distantly supervised relation extraction (DSRE). However, due to relation extraction being performed at bag level, MIL has significant hardware requirements for training when coupled with large sentence encoders such as deep transformer neural networks. In this paper, we propose a novel sampling method for DSRE that relaxes these hardware requirements. In the proposed method, we limit the number of sentences in a batch by randomly sampling sentences from the bags in the batch. However, this comes at the cost of losi…
University of Science and Technology of China（中国科学技术大学）；iFLYTEK Research（科大讯飞研究院）
Cross-language pre-trained models such as multilingual BERT (mBERT) have achieved significant performance in various cross-lingual downstream NLP tasks. This paper proposes a multi-level contrastive learning (ML-CTL) framework to further improve the cross-lingual ability of pre-trained models. The proposed method uses translated parallel data to encourage the model to generate similar semantic embeddings for different languages. However, unlike the sentence-level alignment used in most previous studies, in this paper, we explicitly integrate the word-level information of each pair of parallel …
Tencent（腾讯）；Beijing Institute of Microbiology and Epidemiology（北京微生物与流行病学研究所）；State Key Laboratory of Pathogen and Biosecurity（病原与生物安全国家重点实验室）；National University of Defense Technology（国防科技大学）；National Key Lab of Parallel and Distributed Processing（并行与分布式处理国家重点实验室）
Semantic networks, such as the knowledge graph, can represent the knowledge leveraging the graph structure. Although the knowledge graph shows promising values in natural language processing, it suffers from incompleteness. This paper focuses on knowledge graph completion by predicting linkage between entities, which is a fundamental yet critical task. Semantic matching is a potential solution as it can deal with unseen entities, which the translational distance based methods struggle with. However, to achieve competitive performance as translational distance based methods, semantic matching b…
