Search

Transformer-based pre-training models like BERT have achieved remarkable performance in many natural language processing tasks.However, these models are both computation and memory expensive, hindering their deployment to resource-constrained devices. In this work, we propose TernaryBERT, which ternarizes the weights in a fine-tuned BERT model. Specifically, we use both approximation-based and loss-aware ternarization methods and empirically investigate the ternarization granularity of different parts of BERT. Moreover, to reduce the accuracy degradation caused by the lower capacity of low bits, we leverage the knowledge distillation technique in the training process. Experiments on the GLUE benchmark and SQuAD show that our proposed TernaryBERT outperforms the other BERT quantization methods, and even achieves comparable performance as the full-precision model while being 14.9x smaller., Comment: Accepted by EMNLP 2020

Transformer has been widely-used in many Natural Language Processing (NLP) tasks and the scaled dot-product attention between tokens is a core module of Transformer. This attention is a token-wise design and its complexity is quadratic to the length of sequence, limiting its application potential for long sequence tasks. In this paper, we propose a dimension-wise attention mechanism based on which a novel language modeling approach (namely TensorCoder) can be developed. The dimension-wise attention can reduce the attention complexity from the original $O(N^2d)$ to $O(Nd^2)$, where $N$ is the length of the sequence and $d$ is the dimensionality of head. We verify TensorCoder on two tasks including masked language modeling and neural machine translation. Compared with the original Transformer, TensorCoder not only greatly reduces the calculation of the original model but also obtains improved performance on masked language modeling task (in PTB dataset) and comparable performance on machine translation tasks.

The field of machine translation has progressed tremendously in recent years. Even though the translation quality has improved significantly, current systems are still unable to produce uniformly acceptable machine translations for the variety of possible use cases. In this work, we put machine translation in a cross-lingual pipeline and introduce downstream tasks to define task-specific acceptability of machine translations. This allows us to leverage parallel data to automatically generate acceptability annotations on a large scale, which in turn help to learn acceptability detectors for the downstream tasks. We conduct experiments to demonstrate the effectiveness of our framework for a range of downstream tasks and translation models.

Despite its original goal to jointly learn to align and translate, prior researches suggest that Transformer captures poor word alignments through its attention mechanism. In this paper, we show that attention weights DO capture accurate word alignments and propose two novel word alignment induction methods Shift-Att and Shift-AET. The main idea is to induce alignments at the step when the to-be-aligned target token is the decoder input rather than the decoder output as in previous work. Shift-Att is an interpretation method that induces alignments from the attention weights of Transformer and does not require parameter update or architecture change. Shift-AET extracts alignments from an additional alignment module which is tightly integrated into Transformer and trained in isolation with supervision from symmetrized Shift-Att alignments. Experiments on three publicly available datasets demonstrate that both methods perform better than their corresponding neural baselines and Shift-AET significantly outperforms GIZA++ by 1.4-4.8 AER points., Comment: Accepted by EMNLP2020

By introducing a small set of additional parameters, a probe learns to solve specific linguistic tasks (e.g., dependency parsing) in a supervised manner using feature representations (e.g., contextualized embeddings). The effectiveness of such probing tasks is taken as evidence that the pre-trained model encodes linguistic knowledge. However, this approach of evaluating a language model is undermined by the uncertainty of the amount of knowledge that is learned by the probe itself. Complementary to those works, we propose a parameter-free probing technique for analyzing pre-trained language models (e.g., BERT). Our method does not require direct supervision from the probing tasks, nor do we introduce additional parameters to the probing process. Our experiments on BERT show that syntactic trees recovered from BERT using our method are significantly better than linguistically-uninformed baselines. We further feed the empirically induced dependency structures into a downstream sentiment classification task and find its improvement compatible with or even superior to a human-designed dependency schema., Comment: ACL2020

Masked language model and autoregressive language model are two types of language models. While pretrained masked language models such as BERT overwhelm the line of natural language understanding (NLU) tasks, autoregressive language models such as GPT are especially capable in natural language generation (NLG). In this paper, we propose a probabilistic masking scheme for the masked language model, which we call probabilistically masked language model (PMLM). We implement a specific PMLM with a uniform prior distribution on the masking ratio named u-PMLM. We prove that u-PMLM is equivalent to an autoregressive permutated language model. One main advantage of the model is that it supports text generation in arbitrary order with surprisingly good quality, which could potentially enable new applications over traditional unidirectional generation. Besides, the pretrained u-PMLM also outperforms BERT on a set of downstream NLU tasks., Comment: Accepted by ACL 2020

The pre-trained language models like BERT, though powerful in many natural language processing tasks, are both computation and memory expensive. To alleviate this problem, one approach is to compress them for specific tasks before deployment. However, recent works on BERT compression usually compress the large BERT model to a fixed smaller size. They can not fully satisfy the requirements of different edge devices with various hardware performances. In this paper, we propose a novel dynamic BERT model (abbreviated as DynaBERT), which can flexibly adjust the size and latency by selecting adaptive width and depth. The training process of DynaBERT includes first training a width-adaptive BERT and then allowing both adaptive width and depth, by distilling knowledge from the full-sized model to small sub-networks. Network rewiring is also used to keep the more important attention heads and neurons shared by more sub-networks. Comprehensive experiments under various efficiency constraints demonstrate that our proposed dynamic BERT (or RoBERTa) at its largest size has comparable performance as BERT-base (or RoBERTa-base), while at smaller widths and depths consistently outperforms existing BERT compression methods. Code is available at https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/DynaBERT., Comment: NeurIPS-2020 (Spotlight)

Existing data augmentation approaches for neural machine translation (NMT) have predominantly relied on back-translating in-domain (IND) monolingual corpora. These methods suffer from issues associated with a domain information gap, which leads to translation errors for low frequency and out-of-vocabulary terminology. This paper proposes a dictionary-based data augmentation (DDA) method for cross-domain NMT. DDA synthesizes a domain-specific dictionary with general domain corpora to automatically generate a large-scale pseudo-IND parallel corpus. The generated pseudo-IND data can be used to enhance a general domain trained baseline. The experiments show that the DDA-enhanced NMT models demonstrate consistent significant improvements, outperforming the baseline models by 3.75-11.53 BLEU. The proposed method is also able to further improve the performance of the back-translation based and IND-finetuned NMT models. The improvement is associated with the enhanced domain coverage produced by DDA.

Myocardial infarction (MI) is a scientific term that refers to heart attack. In this study, we infer highly relevant second harmonic generation (SHG) cues from collagen fibers exhibiting highly non-centrosymmetric assembly together with two-photon excited cellular autofluorescence in infarcted mouse heart to quantitatively probe fibrosis, especially targeted at an early stage after MI. We present a robust one-shot machine learning algorithm that enables determination of 2D assembly of collagen with high spatial resolution along with its structural arrangement in heart tissues post-MI with spectral specificity and sensitivity. Detection, evaluation, and precise quantification of fibrosis extent at early stage would guide one to develop treatment therapies that may prevent further progression and determine heart transplant needs for patient survival., Comment: Paper was accepted at the IEEE International Symposium on Biomedical Imaging (ISBI 2020)

Recently, it has been widely accepted by the research community that interactions between humans and cyber-physical infrastructures have played a significant role in determining the performance of the latter. The existing paradigm for designing cyber-physical systems for optimal performance focuses on developing models based on historical data. The impacts of context factors driving human system interaction are challenging and are difficult to capture and replicate in existing design models. As a result, many existing models do not or only partially address those context factors of a new design owing to the lack of capabilities to capture the context factors. This limitation in many existing models often causes performance gaps between predicted and measured results. We envision a new design environment, a cyber-physical human system (CPHS) where decision-making processes for physical infrastructures under design are intelligently connected to distributed resources over cyberinfrastructure such as experiments on design features and empirical evidence from operations of existing instances. The framework combines existing design models with context-aware design-specific data involving human-infrastructure interactions in new designs, using a machine learning approach to create augmented design models with improved predictive powers., Comment: Paper was accepted at the 12th International Conference on Communication Systems and Networks (COMSNETS 2020)

A reverse dictionary takes the description of a target word as input and outputs the target word together with other words that match the description. Existing reverse dictionary methods cannot deal with highly variable input queries and low-frequency target words successfully. Inspired by the description-to-word inference process of humans, we propose the multi-channel reverse dictionary model, which can mitigate the two problems simultaneously. Our model comprises a sentence encoder and multiple predictors. The predictors are expected to identify different characteristics of the target word from the input query. We evaluate our model on English and Chinese datasets including both dictionary definitions and human-written descriptions. Experimental results show that our model achieves the state-of-the-art performance, and even outperforms the most popular commercial reverse dictionary system on the human-written description dataset. We also conduct quantitative analyses and a case study to demonstrate the effectiveness and robustness of our model. All the code and data of this work can be obtained on https://github.com/thunlp/MultiRD., Comment: Accepted by AAAI Conference on Artificial Intelligence 2020

Data modeling and reduction for in situ is important. Feature-driven methods for in situ data analysis and reduction are a priority for future exascale machines as there are currently very few such methods. We investigate a deep-learning based workflow that targets in situ data processing using autoencoders. We propose a Residual Autoencoder integrated Residual in Residual Dense Block (RRDB) to obtain better performance. Our proposed framework compressed our test data into 66 KB from 2.1 MB per 3D volume timestep., Comment: Accepted as a research poster at the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC19)

Story generation is an important natural language processing task that aims to generate coherent stories automatically. While the use of neural networks has proven effective in improving story generation, how to learn to generate an explainable high-level plot still remains a major challenge. In this work, we propose a latent variable model for neural story generation. The model treats an outline, which is a natural language sentence explainable to humans, as a latent variable to represent a high-level plot that bridges the input and output. We adopt an external summarization model to guide the latent variable model to learn how to generate outlines from training data. Experiments show that our approach achieves significant improvements over state-of-the-art methods in both automatic and human evaluations., Comment: 10 pages

Complex node interactions are common in knowledge graphs, and these interactions also contain rich knowledge information. However, traditional methods usually treat a triple as a training unit during the knowledge representation learning (KRL) procedure, neglecting contextualized information of the nodes in knowledge graphs (KGs). We generalize the modeling object to a very general form, which theoretically supports any subgraph extracted from the knowledge graph, and these subgraphs are fed into a novel transformer-based model to learn the knowledge embeddings. To broaden usage scenarios of knowledge, pre-trained language models are utilized to build a model that incorporates the learned knowledge representations. Experimental results demonstrate that our model achieves the state-of-the-art performance on several medical NLP tasks, and improvement above TransE indicates that our KRL method captures the graph contextualized information effectively., Comment: Findings of EMNLP 2020

Reconstruction of seismic data with missing traces is a long-standing issue in seismic data processing. In recent years, rank reduction operations are being commonly utilized to overcome this problem, which require the rank of seismic data to be a prior. However, the rank of field data is unknown; usually it requires much time to manually adjust the rank and just obtain an approximated rank. Methods based on deep learning require very large datasets for training; however acquiring large datasets is difficult owing to physical or financial constraints in practice. Therefore, in this work, we developed a novel method based on unsupervised learning using the intrinsic properties of a convolutional neural network known as U-net, without training datasets. Only one undersampled seismic data was needed, and the deep seismic prior of input data could be exploited by the network itself, thus making the reconstruction convenient. Furthermore, this method can handle both irregular and regular seismic data. Synthetic and field data were tested to assess the performance of the proposed algorithm (DSPRecon algorithm); the advantages of using our method were evaluated by comparing it with the singular spectrum analysis (SSA) method for irregular data reconstruction and de-aliased Cadzow method for regular data reconstruction. Experimental results showed that our method provided better reconstruction performance than the SSA or Cadzow methods. The recovered signal-to-noise ratios (SNRs) were 32.68 dB and 19.11 dB for the DSPRecon and SSA algorithms, respectively. Those for the DSPRecon and Cadzow methods were 35.91 dB and 15.32 dB, respectively., Comment: 5 pages,12 figures

Satellite image classification is a challenging problem that lies at the crossroads of remote sensing, computer vision, and machine learning. Due to the high variability inherent in satellite data, most of the current object classification approaches are not suitable for handling satellite datasets. The progress of satellite image analytics has also been inhibited by the lack of a single labeled high-resolution dataset with multiple class labels. In a preliminary version of this work, we introduced two new high resolution satellite imagery datasets (SAT-4 and SAT-6) and proposed DeepSat framework for classification based on "handcrafted" features and a deep belief network (DBN). The present paper is an extended version, we present an end-to-end framework leveraging an improved architecture that augments a convolutional neural network (CNN) with handcrafted features (instead of using DBN-based architecture) for classification. Our framework, having access to fused spatial information obtained from handcrafted features as well as CNN feature maps, have achieved accuracies of 99.90% and 99.84% respectively, on SAT-4 and SAT-6, surpassing all the other state-of-the-art results. A statistical analysis based on Distribution Separability Criterion substantiates the robustness of our approach in learning better representations for satellite imagery., Comment: This is an Accepted Manuscript of an article published by Taylor & Francis Group in Remote Sensing Letters. arXiv admin note: text overlap with arXiv:1509.03602

Leveraging multilingual parallel texts to automatically generate paraphrases has drawn much attention as size of high-quality paraphrase corpus is limited. Round-trip translation, also known as the pivoting method, is a typical approach to this end. However, we notice that the pivoting process involves multiple machine translation models and is likely to incur semantic drift during the two-step translations. In this paper, inspired by the Transformer-based language models, we propose a simple and unified paraphrasing model, which is purely trained on multilingual parallel data and can conduct zero-shot paraphrase generation in one step. Compared with the pivoting approach, paraphrases generated by our model is more semantically similar to the input sentence. Moreover, since our model shares the same architecture as GPT (Radford et al., 2018), we are able to pre-train the model on large-scale unparallel corpus, which further improves the fluency of the output sentences. In addition, we introduce the mechanism of denoising auto-encoder (DAE) to improve diversity and robustness of the model. Experimental results show that our model surpasses the pivoting method in terms of relevance, diversity, fluency and efficiency.

Despite the success of neural machine translation (NMT), simultaneous neural machine translation (SNMT), the task of translating in real time before a full sentence has been observed, remains challenging due to the syntactic structure difference and simultaneity requirements. In this paper, we propose a general framework for adapting neural machine translation to translate simultaneously. Our framework contains two parts: prefix translation that utilizes a consecutive NMT model to translate source prefixes and a stopping criterion that determines when to stop the prefix translation. Experiments on three translation corpora and two language pairs show the efficacy of the proposed framework on balancing the quality and latency in adapting NMT to perform simultaneous translation., Comment: Accepted by AACL 2020

Previous work on document-level NMT usually focuses on limited contexts because of degraded performance on larger contexts. In this paper, we investigate on using large contexts with three main contributions: (1) Different from previous work which pertrained models on large-scale sentence-level parallel corpora, we use pretrained language models, specifically BERT, which are trained on monolingual documents; (2) We propose context manipulation methods to control the influence of large contexts, which lead to comparable results on systems using small and large contexts; (3) We introduce a multi-task training for regularization to avoid models overfitting our training corpora, which further improves our systems together with a deeper encoder. Experiments are conducted on the widely used IWSLT data sets with three language pairs, i.e., Chinese--English, French--English and Spanish--English. Results show that our systems are significantly better than three previously reported document-level systems.

Adversarial attacks are carried out to reveal the vulnerability of deep neural networks. Textual adversarial attacking is challenging because text is discrete and a small perturbation can bring significant change to the original input. Word-level attacking, which can be regarded as a combinatorial optimization problem, is a well-studied class of textual attack methods. However, existing word-level attack models are far from perfect, largely because unsuitable search space reduction methods and inefficient optimization algorithms are employed. In this paper, we propose a novel attack model, which incorporates the sememe-based word substitution method and particle swarm optimization-based search algorithm to solve the two problems separately. We conduct exhaustive experiments to evaluate our attack model by attacking BiLSTM and BERT on three benchmark datasets. Experimental results demonstrate that our model consistently achieves much higher attack success rates and crafts more high-quality adversarial examples as compared to baseline methods. Also, further experiments show our model has higher transferability and can bring more robustness enhancement to victim models by adversarial training. All the code and data of this paper can be obtained on https://github.com/thunlp/SememePSO-Attack., Comment: Accepted at ACL 2020 as a long paper (a typo is corrected as compared with the official conference camera-ready version). 16 pages, 3 figures

Sememes, the minimum semantic units of human languages, have been successfully utilized in various natural language processing applications. However, most existing studies exploit sememes in specific tasks and few efforts are made to utilize sememes more fundamentally. In this paper, we propose to incorporate sememes into recurrent neural networks (RNNs) to improve their sequence modeling ability, which is beneficial to all kinds of downstream tasks. We design three different sememe incorporation methods and employ them in typical RNNs including LSTM, GRU and their bidirectional variants. In evaluation, we use several benchmark datasets involving PTB and WikiText-2 for language modeling, SNLI for natural language inference and another two datasets for sentiment analysis and paraphrase detection. Experimental results show evident and consistent improvement of our sememe-incorporated models compared with vanilla RNNs, which proves the effectiveness of our sememe incorporation methods. Moreover, we find the sememe-incorporated models have higher robustness and outperform adversarial training in defending adversarial attack. All the code and data of this work can be obtained at https://github.com/thunlp/SememeRNN., Comment: Published in IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP). 10 pages, 2 figures

Language model pre-training, such as BERT, has significantly improved the performances of many natural language processing tasks. However, pre-trained language models are usually computationally expensive, so it is difficult to efficiently execute them on resource-restricted devices. To accelerate inference and reduce model size while maintaining accuracy, we first propose a novel Transformer distillation method that is specially designed for knowledge distillation (KD) of the Transformer-based models. By leveraging this new KD method, the plenty of knowledge encoded in a large teacher BERT can be effectively transferred to a small student Tiny-BERT. Then, we introduce a new two-stage learning framework for TinyBERT, which performs Transformer distillation at both the pretraining and task-specific learning stages. This framework ensures that TinyBERT can capture he general-domain as well as the task-specific knowledge in BERT. TinyBERT with 4 layers is empirically effective and achieves more than 96.8% the performance of its teacher BERTBASE on GLUE benchmark, while being 7.5x smaller and 9.4x faster on inference. TinyBERT with 4 layers is also significantly better than 4-layer state-of-the-art baselines on BERT distillation, with only about 28% parameters and about 31% inference time of them. Moreover, TinyBERT with 6 layers performs on-par with its teacher BERTBASE., Comment: Findings of EMNLP 2020; results have been updated; code and model: https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/TinyBERT

The pre-trained language models have achieved great successes in various natural language understanding (NLU) tasks due to its capacity to capture the deep contextualized information in text by pre-training on large-scale corpora. In this technical report, we present our practice of pre-training language models named NEZHA (NEural contextualiZed representation for CHinese lAnguage understanding) on Chinese corpora and finetuning for the Chinese NLU tasks. The current version of NEZHA is based on BERT with a collection of proven improvements, which include Functional Relative Positional Encoding as an effective positional encoding scheme, Whole Word Masking strategy, Mixed Precision Training and the LAMB Optimizer in training the models. The experimental results show that NEZHA achieves the state-of-the-art performances when finetuned on several representative Chinese tasks, including named entity recognition (People's Daily NER), sentence matching (LCQMC), Chinese sentiment classification (ChnSenti) and natural language inference (XNLI).

Neural dialog state trackers are generally limited due to the lack of quantity and diversity of annotated training data. In this paper, we address this difficulty by proposing a reinforcement learning (RL) based framework for data augmentation that can generate high-quality data to improve the neural state tracker. Specifically, we introduce a novel contextual bandit generator to learn fine-grained augmentation policies that can generate new effective instances by choosing suitable replacements for the specific context. Moreover, by alternately learning between the generator and the state tracker, we can keep refining the generative policies to generate more high-quality training data for neural state tracker. Experimental results on the WoZ and MultiWoZ (restaurant) datasets demonstrate that the proposed framework significantly improves the performance over the state-of-the-art models, especially with limited training data., Comment: AAAI 2020

Due to the sparsity of features, noise has proven to be a great inhibitor in the classification of handwritten characters. To combat this, most techniques perform denoising of the data before classification. In this paper, we consolidate the approach by training an all-in-one model that is able to classify even noisy characters. For classification, we progressively train a classifier generative adversarial network on the characters from low to high resolution. We show that by learning the features at each resolution independently a trained model is able to accurately classify characters even in the presence of noise. We experimentally demonstrate the effectiveness of our approach by classifying noisy versions of MNIST, handwritten Bangla Numeral, and Basic Character datasets., Comment: Paper was accepted at the 21st International Conference on Asia-Pacific Digital Libraries (ICADL 2019)

Semantic compositionality (SC) refers to the phenomenon that the meaning of a complex linguistic unit can be composed of the meanings of its constituents. Most related works focus on using complicated compositionality functions to model SC while few works consider external knowledge in models. In this paper, we verify the effectiveness of sememes, the minimum semantic units of human languages, in modeling SC by a confirmatory experiment. Furthermore, we make the first attempt to incorporate sememe knowledge into SC models, and employ the sememeincorporated models in learning representations of multiword expressions, a typical task of SC. In experiments, we implement our models by incorporating knowledge from a famous sememe knowledge base HowNet and perform both intrinsic and extrinsic evaluations. Experimental results show that our models achieve significant performance boost as compared to the baseline methods without considering sememe knowledge. We further conduct quantitative analysis and case studies to demonstrate the effectiveness of applying sememe knowledge in modeling SC. All the code and data of this paper can be obtained on https://github.com/thunlp/Sememe-SC., Comment: To appear at ACL 2019

We present a simple yet effective method for generating high quality classical Chinese poetry with Generative Pre-trained Language Model (GPT). The method adopts a simple GPT model, without using any human crafted rules or features, or designing any additional neural components. While the proposed model learns to generate various forms of classical Chinese poems, including Jueju, L\"{u}shi, various Cipai and Couples, the generated poems are of very high quality. We also propose and implement a method to fine-tune the model to generate acrostic poetry. To the best of our knowledge, this is the first to employ GPT in developing a poetry generation system. We have released an online mini demonstration program on Wechat to show the generation capability of the proposed method for classical Chinese poetry.