Search

Insects maintain remarkable agility after incurring severe injuries or wounds. Although robots driven by rigid actuators have demonstrated agile locomotion and manipulation, most of them lack animal-like robustness against unexpected damage. Dielectric elastomer actuators (DEAs) are a class of muscle-like soft transducers that have enabled nimble aerial, terrestrial, and aquatic robotic locomotion comparable to that of rigid actuators. However, unlike muscles, DEAs suffer local dielectric breakdowns that often cause global device failure. These local defects severely limit DEA performance, lifetime, and size scalability. We developed DEAs that can endure more than 100 punctures while maintaining high bandwidth (>400 hertz) and power density (>700 watt per kilogram)—sufficient for supporting energetically expensive locomotion such as flight. We fabricated electroluminescent DEAs for visualizing electrode connectivity under actuator damage. When the DEA suffered severe dielectric breakdowns that caused device failure, we demonstrated a laser-assisted repair method for isolating the critical defects and recovering performance. These results culminate in an aerial robot that can endure critical actuator and wing damage while maintaining similar accuracy in hovering flight. Our work highlights that soft robotic systems can embody animal-like agility and resilience—a critical biomimetic capability for future robots to interact with challenging environments.

The hypotactic structural relation between clauses plays an important role in improving the discourse coherence of document-level translation. However, the standard neural machine translation (NMT) models do not explicitly model the hypotactic relationship between clauses, which usually leads to structurally incorrect translations of long and complex sentences. This problem is particularly noticeable on Chinese-to-English translation task of complex sentences due to the grammatical form distinction between English and Chinese. English is rich in grammatical form (e.g. verb morphological changes and subordinating conjunctions) while Chinese is poor in grammatical form. These linguistic phenomena make it a challenge for NMT to learn the hypotactic structure knowledge from Chinese as well as the structure alignment between Chinese and English. To address these issues, we propose to model the hypotactic structure for Chinese-to-English complex sentence translation by introducing hypotactic structure knowledge. Specifically, we annotate and build a hypotactic structure aligned parallel corpus that provides rich hypotactic structure knowledge for NMT. Moreover, we further propose a structure-infused neural framework to combine the hypotactic structure knowledge with the NMT model through two integrating strategies. In particular, we introduce a specific structure-aware loss to encourage the NMT model to better learn the structure knowledge. Experimental results on WMT17, WMT18 and WMT19 Chinese-to-English translation tasks demonstrate the effectiveness of the proposed methods.

Decision fusion is an effective way to resolve the conflict of diagnosis results. Aiming at the problem that Dempster-Shafer (DS) theory deals with the high conflict of evidence and produces wrong results, a decision fusion algorithm for fault diagnosis based on closeness and DS theory is proposed. Firstly, the relevant concepts of DS theory are introduced, and the normal distribution membership function is used as the evidence closeness. Secondly, the harmonic average is introduced, and the weight of each evidence is established according to the product of closeness of each evidence and its harmonic average. Thirdly, the weight of conflicting evidence is regularized, and the final decision fusion result is obtained by using the Dempster’s rule. Lastly, the simulation and application examples are designed. Simulation and application results show that the method can effectively reduce the impact of diagnostic information conflicts and improve the accuracy of decision fusion. What’s more, the method considers the overall average distribution of evidence in the identification framework, it can reduce evidence conflicts while preserving important diagnostic information.

Millimeter-sized electrostatic film actuators, inspired by the efficient spatial arrangement of insect muscles, achieve a muscle-like power density (61 W kg−1) and enable robotic applications in which agility is needed in confined spaces. Like biological muscles, these actuators incorporate a hierarchical structure, in this case building from electrodes to arrays to laminates, and are composed primarily of flexible materials. So comprised, these actuators can be designed for a wide range of manipulation and locomotion tasks, similar to natural muscle, while being robust and compact. A typical actuator can achieve 85 mN of force with a 15 mm stroke, with a size of [Formula: see text] mm3 and mass of 92 mg. Two millimeter-sized robots, an ultra-thin earthworm-inspired robot and an intestinal-muscle-inspired endoscopic tool for tissue resection, demonstrate the utility of these actuators. The earthworm robot undertakes inspection tasks: the navigation of a 5 mm channel and a 19 mm square tube while carrying an on-board camera. The surgical tool, which conforms to the surface of the distal end of an endoscope, similar to the thin, smooth muscle that covers the intestine, completes tissue cutting and penetrating tasks. Beyond these devices, we anticipate widespread use of these actuators in soft robots, medical robots, wearable robots, and miniature autonomous systems.

The goal of sentence matching is to determine the semantic relation between two sentences, which is the basis of many downstream tasks in natural language processing, such as question answering and information retrieval. Recent studies using attention mechanism to align the elements of two sentences have shown promising results in capturing semantic similarity/relevance. Most existing methods mainly focus on the design of multi-layer attention network, however, some critical issues have not been dealt with well: 1) the higher attention layer is easily affected by error propagation because it relies on the alignment results of preceding attentions; 2) models have the risk of losing low-layer semantic features with the increase of network depth; and 3) the approach of capturing global matching information brings about large computing complexity for model training. To this end, we propose a Deep Bi-Directional Interaction Network (DBDIN) to solve these issues, which captures semantic relatedness from two directions and each direction employs multiple attention-based interaction units. To be specific, the attention of each interaction unit will repeatedly focus on the original sentence representation of another one for semantic alignment, which alleviates the error propagation problem by attending to a fixed semantic representation. Then we design deep fusion to aggregate and propagate attention information from low layers to high layers, which effectively retains low-layer semantic features for subsequential interactions. Moreover, we introduce a self-attention mechanism at last to enhance global matching information with smaller model complexity. We conduct experiments on natural language inference and paraphrase identification tasks with three benchmark datasets SNLI, SciTail and Quora. Experimental results demonstrate that our proposed method can achieve significant improvements over baseline systems without using any external knowledge. Additionally, we conduct interpretable study to disclose how our deep interaction network with attention can benefit sentence matching, which provides a reference for future model design. Ablation studies and visualization analyses further verify that our model can better capture interactive information between two sentences, and the proposed components are indeed able to help modeling semantic relation more precisely.

SUMMARYIn this paper, convolutional neural networks (CNNs) were used to distinguish between tectonic and non-tectonic seismicity. The proposed CNNs consisted of seven convolutional layers with small kernels and one fully connected layer, which only relied on the acoustic waveform without extracting features manually. For a single station, the accuracy of the model was 0.90, and the event accuracy could reach 0.93. The proposed model was tested using data from January 2019 to August 2019 in China. The event accuracy could reach 0.92, showing that the proposed model could distinguish between tectonic and non-tectonic seismicity.

Many insects demonstrate remarkable locomotive capabilities on inclined or even inverted surfaces. Achieving inverted locomotion is a challenge for legged insect-scale robots because repeated attachment and detachment to a surface usually requires the design of special climbing gaits, adhesion mechanisms, sensing, and feedback control. In this study, we propose a novel adhesion method that leverages capillary and lubrication effects to achieve simultaneous adhesion and sliding. We design a 47 mg adhesion pad and install it on a 1.4 g insect-scale quadrupedal robot to demonstrate locomotion on inverted and inclined surfaces. On an inverted acrylic surface, the robot's climbing and turning speeds are 0.3 cm/s and 23.6 °/s, respectively. Further, the robot can climb a 30° inclined acrylic surface at 0.04 cm/s. This light-weight, passively stable, and versatile adhesion design is suitable for insect-scale robots with limited sensing, actuation, and control capabilities.

Tongue coating has been used as an effective signature of health in traditional Chinese medicine (TCM). The level of greasy coating closely relates to the strength of dampness or pathogenic qi in TCM theory. Previous empirical studies and our systematic review have shown the relation between greasy coating and various diseases, including gastroenteropathy, coronary heart disease, and coronavirus disease 2019 (COVID-19). However, the objective and intelligent greasy coating and related diseases recognition methods are still lacking. The construction of the artificial intelligent tongue recognition models may provide important syndrome diagnosis and efficacy evaluation methods, and contribute to the understanding of ethnopharmacological mechanisms based on TCM theory.The present study aimed to develop an artificial intelligent model for greasy tongue coating recognition and explore its application in COVID-19.Herein, we developed greasy tongue coating recognition networks (GreasyCoatNet) using convolutional neural network technique and a relatively large (N = 1486) set of tongue images from standard devices. Tests were performed using both cross-validation procedures and a new dataset (N = 50) captured by common cameras. Besides, the accuracy and time efficiency comparisons between the GreasyCoatNet and doctors were also conducted. Finally, the model was transferred to recognize the greasy coating level of COVID-19.The overall accuracy in 3-level greasy coating classification with cross-validation was 88.8% and accuracy on new dataset was 82.0%, indicating that GreasyCoatNet can obtain robust greasy coating estimates from diverse datasets. In addition, we conducted user study to confirm that our GreasyCoatNet outperforms TCM practitioners, yet only consuming roughly 1% of doctors' examination time. Critically, we demonstrated that GreasyCoatNet, along with transfer learning, can construct more proper classifier of COVID-19, compared to directly training classifier on patient versus control datasets. We, therefore, derived a disease-specific deep learning network by finetuning the generic GreasyCoatNet.Our framework may provide an important research paradigm for differentiating tongue characteristics, diagnosing TCM syndrome, tracking disease progression, and evaluating intervention efficacy, exhibiting its unique potential in clinical applications.

With task-oriented dialogue systems being widely applied in everyday life, slot filling, the essential component of task-oriented dialogue systems, is required to be quickly adapted to new domains that contain domain-specific slots with few or no training data. Previous methods for slot filling usually adopt sequence labeling framework, which, however, often has limited ability when dealing with the domain-specific slots. In this paper, we take a new perspective on cross-domain slot filling by framing it as a machine reading comprehension (MRC) problem. Our approach firstly transforms slot names into well-designed queries, which contain rich informative prior knowledge and are very helpful for the detection of domain-specific slots. In addition, we utilize the large-scale MRC dataset for pre-training, which further alleviates the data scarcity problem. Experimental results on SNIPS and ATIS datasets show that our approach consistently outperforms the existing state-of-the-art methods by a large margin.

Stylized neural machine translation (NMT) aims to translate sentences of one style into sentences of another style, which is essential for the application of machine translation in a real-world scenario. However, a major challenge in this task is the scarcity of high-quality parallel data which is stylized paired. To address this problem, we propose an iterative dual knowledge transfer framework that utilizes informal training data of machine translation and formality style transfer data to create large-scale stylized paired data, for the training of stylized machine translation model. Specifically, we perform bidirectional knowledge transfer between translation model and text style transfer model iteratively through knowledge distillation. Then, we further propose a data-refinement module to process the noisy synthetic parallel data generated during knowledge transfer. Experiment results demonstrate the effectiveness of our method, achieving an improvement over the existing best model by 5 BLEU points on MTFC dataset. Meanwhile, extensive analyses illustrate our method can also improve the accuracy of formality style transfer.

Aiming at the dimensional disaster and over-fitting problems in data analysis, this paper proposes a feature selection method using hybrid integration of difference models and random forests (Integrate-RF), firstly, Integrate-RF use CART, CHAID, SVM, BN, NN, K-Means, Kohonen to evaluate the importance of features, and then, for the above seven sorts, Integrate-RF use the arithmetic average method to calculate the importance of the features; secondly, Integrate-RF select the most important features from the remaining features into features subset, and use random forest classification to get the corresponding out-of-bag(OOB) data classification error rate; finally, the optimal features subset can be selected based on the OOB data classification error rate. Experiments show that feature selection methods proposed in this paper effectively reduces the data dimension, selects features better and more adaptable.

Scheduled sampling is an effective method to alleviate the exposure bias problem of neural machine translation. It simulates the inference scene by randomly replacing ground-truth target input tokens with predicted ones during training. Despite its success, its critical schedule strategies are merely based on training steps, ignoring the real-time model competence, which limits its potential performance and convergence speed. To address this issue, we propose confidence-aware scheduled sampling. Specifically, we quantify real-time model competence by the confidence of model predictions, based on which we design fine-grained schedule strategies. In this way, the model is exactly exposed to predicted tokens for high-confidence positions and still ground-truth tokens for low-confidence positions. Moreover, we observe vanilla scheduled sampling suffers from degenerating into the original teacher forcing mode since most predicted tokens are the same as ground-truth tokens. Therefore, under the above confidence-aware strategy, we further expose more noisy tokens (e.g., wordy and incorrect word order) instead of predicted ones for high-confidence token positions. We evaluate our approach on the Transformer and conduct experiments on large-scale WMT 2014 English-German, WMT 2014 English-French, and WMT 2019 Chinese-English. Results show that our approach significantly outperforms the Transformer and vanilla scheduled sampling on both translation quality and convergence speed., Findings of ACL-2021, code at https://github.com/Adaxry/conf_aware_ss4nmt

Distributed Raman amplifier (DRA) has been widely studied in recent decades because of its low noise figure and flexible gain. In this paper, we present a novel scheme of DRA with broadband amplified spontaneous emission(ASE) source as pump instead of discrete pump lasers. The broadband pump is optimized by machine learning based inverse design and shaped by programmable waveshaper, so as to realize the ultrafine, dynamic and arbitrary gain spectrum shaping of Raman amplification. For the target of flat gain spectrum, the maximum gain flatness of 0.1086 dB is realized based on the simulation results. For the target of arbitrary gain spectrum, we demonstrate four gain profiles with maximum root mean square error (RMSE) of 0.074 dB. To further measure the performance of arbitrary gain spectrum optimization, the probability density functions (PDF) of RMSE and Errormax are presented. Meanwhile, the numeral relationship between the bands of broadband pump and signal is also explored. Furthermore, this work has great application potential to compensate the gain distortion or dynamic change caused by other devices in communication systems.

This article addresses the forbidden state problem in discrete-event systems (DESs) modeled with Petri nets. Given a control specification, we first decide the sets of forbidden and admissible markings. Then, the minimal mask set of first-met forbidden markings (FFMs) and the minimal root set of admissible markings are computed by marking mask that is implemented using a class of special places in a plant, called competitive places. Marking mask can effectively filtrate the markings to be processed such that the two obtained sets are in general much smaller than the sets of originally specified forbidden and admissible markings, respectively. Monitors computed by place invariants are used to forbid the forbidden markings. It is shown that a maximally permissive (optimal) supervisor can be computed if it exists. Integer linear programming is used to optimize the structure of a supervisor. The minimal mask set of FFMs and root set of admissible markings efficiently reduce the computational overhead because of much fewer constraints and variables in the formulated programming problem. The developed methodology is illustrated by parameterized examples.

Flying insects capable of navigating in highly cluttered natural environments can withstand in-flight collisions because of the combination of their low inertia1 and the resilience of their wings2, exoskeletons1 and muscles. Current insect-scale (less than ten centimetres long and weighing less than five grams) aerial robots3–6 use rigid microscale actuators, which are typically fragile under external impact. Biomimetic artificial muscles7–10 that are capable of large deformation offer a promising alternative for actuation because they can endure the stresses caused by such impacts. However, existing soft actuators11–13 have not yet demonstrated sufficient power density to achieve lift-off, and their actuation nonlinearity and limited bandwidth create further challenges for achieving closed-loop (driven by an input control signal that is adjusted based on sensory feedback) flight control. Here we develop heavier-than-air aerial robots powered by soft artificial muscles that demonstrate open-loop (driven by a predetermined signal without feedback), passively stable (upright during flight) ascending flight as well as closed-loop, hovering flight. The robots are driven by multi-layered dielectric elastomer actuators that weigh 100 milligrams each and have a resonance frequency of 500 hertz and power density of 600 watts per kilogram. To increase the mechanical power output of the actuator and to demonstrate flight control, we present ways to overcome challenges unique to soft actuators, such as nonlinear transduction and dynamic buckling. These robots can sense and withstand collisions with surrounding obstacles and can recover from in-flight collisions by exploiting material robustness and vehicle passive stability. We also fly two micro-aerial vehicles simultaneously in a cluttered environment. They collide with the wall and each other without suffering damage. These robots rely on offboard amplifiers and an external motion-capture system to provide power to the dielectric elastomer actuators and to control their flight. Our work demonstrates how soft actuators can achieve sufficient power density and bandwidth to enable controlled flight, illustrating the potential of developing next-generation agile soft robots. Heavier-than-air insect-scale aerial robots powered by soft artificial muscles can hover and also recover from in-flight collisions, illustrating the potential for developing next-generation agile soft robots.

In this work, we proposed a long-term tracking strategy to deal with the occlusion, out-of-plane rotation, and the confusing non-target object. Our tracking system is composed of two parts, the CA-CF tracker, an efficient correlation method for short-term tracking, and the SVM-based re-detector, which prevents the CA tracker from degradation. When the tracker works with confidence, the CA-CF module ensures an accurate tracking result and the SVM updates accordingly. When the response maps fluctuate heavily, the SVM switches to work as a re-detector and the tracker will be initialized. We also introduced to adopt both the maximum response criterion and the APCE criterion to judge the performance of the tracker in time. By evaluating our algorithm on the OTB benchmark datasets, we proposed to analyze the result affected by the parameters of our CA-CF-SVM strategy. The experimental results show that our method has a significant improvement than the state-of-the-art methods for the long-term tracking both in accuracy and robustness.

Hyperspectral imaging has great achievements in agriculture, astronomy, surveillance, and so on. However, the inherent low spatial resolution of hyperspectral imaging, unfortunately, limits its more widespread applications. Recently, hyperspectral image (HSI) super resolution addresses this problem by fusing a low spatial resolution HSI (LR-HSI) with a high spatial resolution multispectral image (HR-MSI), but most of these methods did not consider real-time restoration of high spatial resolution HSI. In this study, the authors propose a fast HSI super-resolution method which fills this blank. Specifically, they model the hyperspectral super resolution as a linear regression problem according to the fact that the imaging process is a linear transform and the inverse of this transform can be approximately estimated, as the spectra of a typical scene lie in a very low-dimensional space. To further exploit the low-dimensional nature of the spectra, they divide the HR-MSI and LR-HSI into several patches and learn the inverse transform patch-by-patch. Experiments on several public datasets show that their method approximates state-of-the-art methods in accuracy, but is several orders of magnitude faster than all of them. Furthermore, they provide an efficient C language implementation of their methods, which can meet the real-time request.

It is difficult to find the hidden danger or defect of substation equipment accurately and quickly when the substation equipment is inspected manually. In this paper, an electric inspection robot equipped with ultraviolet camera was studied. The dual-light image detection module of visible and ultraviolet camera was built on the robot's head. The effect of air pressure on the results of ultraviolet imaging was studied experimentally. The results show that with the increase of air pressure, the spot area decreases obviously, and the corona discharge weakens. Robot patrol inspection equipped with ultraviolet camera is helpful to realize non-contact live detection, which is of great significance to improve the intelligent detection level of substation.

Node.js is one of the most popular frameworks for building web applications. As software systems mature, the cost of running their entire regression test suite can become significant. Selective Regression Testing (SRT) is a technique that executes only a subset of tests the regression test suite can detect software failures more efficiently. Previous SRT studies mainly focused on standard desktop applications. Node.js applications are considered hard to perform test reduction because of Node's asynchronous, event-driven programming model and because JavaScript is a dynamic programming language. In this paper, we present NodeSRT, a Selective Regression Testing framework for Node.js applications. By performing static and dynamic analysis, NodeSRT identifies the relationship between changed methods and tests, then reduces the regression test suite to only tests that are affected by the change to improve the execution time of the regression test suite. To evaluate our selection technique, we applied NodeSRT to two open-source projects: Uppy and Simorgh, then compared our approach with the retest-all strategy and current industry-standard SRT technique: Jest OnlyChange. The results demonstrate that NodeSRT correctly selects affected tests based on changes and is 250% faster, 450% more precise than the Jest OnlyChange.

Neural chat translation aims to translate bilingual conversational text, which has a broad application in international exchanges and cooperation. Despite the impressive performance of sentence-level and context-aware Neural Machine Translation (NMT), there still remain challenges to translate bilingual conversational text due to its inherent characteristics such as role preference, dialogue coherence, and translation consistency. In this paper, we aim to promote the translation quality of conversational text by modeling the above properties. Specifically, we design three latent variational modules to learn the distributions of bilingual conversational characteristics. Through sampling from these learned distributions, the latent variables, tailored for role preference, dialogue coherence, and translation consistency, are incorporated into the NMT model for better translation. We evaluate our approach on the benchmark dataset BConTrasT (English-German) and a self-collected bilingual dialogue corpus, named BMELD (English-Chinese). Extensive experiments show that our approach notably boosts the performance over strong baselines by a large margin and significantly surpasses some state-of-the-art context-aware NMT models in terms of BLEU and TER. Additionally, we make the BMELD dataset publicly available for the research community., Comment: Accepted as a long paper at ACL 2021. Code and data are available at https://github.com/XL2248/CPCC

Zero-resource named entity recognition (NER) severely suffers from data scarcity in a specific domain or language. Most studies on zero-resource NER transfer knowledge from various data by fine-tuning on different auxiliary tasks. However, how to properly select training data and fine-tuning tasks is still an open problem. In this paper, we tackle the problem by transferring knowledge from three aspects, i.e., domain, language and task, and strengthening connections among them. Specifically, we propose four practical guidelines to guide knowledge transfer and task fine-tuning. Based on these guidelines, we design a target-oriented fine-tuning (TOF) framework to exploit various data from three aspects in a unified training manner. Experimental results on six benchmarks show that our method yields consistent improvements over baselines in both cross-domain and cross-lingual scenarios. Particularly, we achieve new state-of-the-art performance on five benchmarks., Comment: 9 pages, ACL21 Findings

The predominant approach of visual question answering (VQA) relies on encoding the image and question with a “black box" neural encoder and decoding a single token into answers such as “yes” or “no”. Despite this approach’s strong quantitative results, it struggles to come up with human-readable forms of justification for the prediction process. To address this insufficiency, we propose LRRA [Look, Read, Reasoning,Answer], a transparent neural-symbolic framework for visual question answering that solves the complicated problem in the real world step-by-step like humans and provides human-readable form of justification at each step. Specifically, LRRA learns to first convert an image into a scene graph and parse a question into multiple reasoning instructions. It then executes the reasoning instructions one at a time by traversing the scene graph using a recurrent neural-symbolic execution module. Finally, it generates answers to the given questions and makes corresponding marks on the image. Furthermore, we believe that the relations between objects in the question is of great significance for obtaining the correct answer, so we create a perturbed GQA test set by removing linguistic cues (attributes and relations) in the questions to analyze which part of the question contributes more to the answer. Our experiments on the GQA dataset show that LRRA is significantly better than the existing representative model (57.12% vs. 56.39%). Our experiments on the perturbed GQA test set show that the relations between objects is more important for answering complicated questions than the attributes of objects.

Entity Linking (EL) refers to the task of linking entity mentions in the text to the correct entities in the Knowledge Base (KB) in which entity embeddings play a vital and challenging role because of the subtle differences between entities. However, existing pre-trained entity embeddings only learn the underlying semantic information in texts, yet the fine-grained entity type information is ignored, which causes the type of the linked entity is incompatible with the mention context. In order to solve this problem, we propose to encode fine-grained type information into entity embeddings. We firstly pre-train word vectors to inject type information by embedding words and fine-grained entity types into the same vector space. Then we retrain entity embeddings with word vectors containing fine-grained type information. By applying our entity embeddings to two existing EL models, our method respectively achieves 0.82\(\%\) and 0.42\(\%\) improvement on average F1 score of the test sets. Meanwhile, our method is model-irrelevant, which means it can help other EL models.

This paper presents the systems developed by Beijing Jiaotong University and Toshiba (China) Co., Ltd. for the CCMT 2021 quality estimation (QE) and automatic-post editing (APE) task. For QE task, we mainly rely on multiple pretrained language models, and propose a multi-phase pre-finetuning scheme, to adapt the pretrained models to the target domain and task. The pre-finetuning scheme consists of language-adaptative finetuning, domain-adaptative finetuning and task-adaptative finetuning. For APE task, we use BERT-initialized Transformer as the backbone model, and create different groups of synthetic data by different data augmentation methods, i.e. forward translation, round-trip translation and multi-source denoising autoencoder. Multi-model ensemble is adopted in both tasks. Experiment results on the development set show high accuracy on both QE and APE tasks, demonstrating the effectiveness of our proposed methods.

Syntactically controlled paraphrase generation can produce diverse paraphrases by exposing syntactic control, where both semantic preservation and syntactic variations are two important factors. Previous works mainly focus on using fine-grained syntactic structures (e.g., full parse tree) as syntactic control. While these methods can achieve excellent syntactic controllability, leads to failing to preserve the semantics of the input sentence. The main reason is that it is difficult to retrieve perfectly compatible syntactic structures with the input sentences. In this paper, we explore coarse-grained syntactic structures to trade-off semantic preservation and syntactic variations. Furthermore, to improve semantic preservation and syntactic controllability, we propose a Syntax Attention-Guided Paraphrase (SAGP) model that can correctly select syntactic information according to the current state for surface realization. Experiment results show that SAGP outperforms the previous state-of-the-art method under the same setting. Additionally, we validate that using coarse-grained structures can generate more semantically reasonable text without affecting the syntactic controllability.

Deep neural networks have achieved state-of-the-art performances on named entity recognition (NER) with sufficient training data, while they perform poorly in low-resource scenarios due to data scarcity. To solve this problem, we propose a novel data augmentation method based on pre-trained language model (PLM) and curriculum learning strategy. Concretely, we use the PLM to generate diverse training instances through predicting different masked words and design a task-specific curriculum learning strategy to alleviate the influence of noises. We evaluate the effectiveness of our approach on three datasets: CoNLL-2003, OntoNotes5.0, and MaScip, of which the first two are simulated low-resource scenarios, and the last one is a real low-resource dataset in material science domain. Experimental results show that our method consistently outperform the baseline model. Specifically, our method achieves an absolute improvement of 3.46% \(F_1\) score on the 1% CoNLL-2003, 2.58% on the 1% OntoNotes5.0, and 0.99% on the full of MaScip.

Machine translation quality estimation (QE) aims to evaluate the result of translation without reference. Existing approaches require large amounts of training data or model-related features, leading to impractical applications in real world. In this work, we propose a contrastive learning framework to train QE model with limited parallel data. Concretely, we use denoising autoencoder to create negative samples based on sentence reconstruction. Then the QE model is trained to distinguish the golden pair from the negative samples in a contrastive manner. To this end, we propose two contrastive learning architectures, namely Contrastive Classification and Contrastive Ranking. Experiments on four language pairs of MLQE dataset show that our method achieves strong results in both zero-shot and supervised settings. To the best of our knowledge, this is the first trial of contrastive learning on QE.

This paper presents the systems developed by Beijing Jiaotong University for the CCMT2021 evaluation tasks. We joined four translation tasks of Chinese-English, English-Chinese, Uyghur-Chinese, Tibetan-Chinese. In all directions, we build our system based on transformer architecture and Dynamic-Conv. Additionally, we apply Byte Pair Encoding (BPE) to all translation tasks to resolve the out-of-vocabulary (OOV) problem. We also adopt some techniques that have been proven effective recently in academia, such as data augmentation, finetuning, model ensemble and reranking. Experiments show that our machine translation systems achieved high accuracy on all directions.

The successful emotional conversation system depends on sufficient perception and appropriate expression of emotions. In a real-life conversation, humans firstly instinctively perceive emotions from multi-source information, including the emotion flow hidden in dialogue history, facial expressions, audio, and personalities of speakers. Then, they convey suitable emotions according to their personalities, but these multiple types of information are insufficiently exploited in emotional conversation fields. To address this issue, in this paper, we propose a heterogeneous graph-based model for emotional conversation generation. Firstly, we design a Heterogeneous Graph-Based Encoder to represent the conversation content (i.e., the dialogue history, its emotion flow, facial expressions, audio, and speakers' personalities) with a heterogeneous graph neural network, and then predict suitable emotions for feedback. Secondly, we employ an Emotion-Personality-Aware Decoder to generate a response relevant to the conversation context as well as with appropriate emotions, through taking the encoded graph representations, the predicted emotions by the encoder and the personality of the current speaker as inputs. Experiments on both automatic and human evaluation show that our method can effectively perceive emotions from multi-source knowledge and generate a satisfactory response. Furthermore, based on the up-to-date text generator BART, our model still can achieve consistent improvement, which significantly outperforms some existing state-of-the-art models., Accepted at Artificial Intelligence as an extension of the AAAI conference version

Paraphrase generation is of great importance to many downstream tasks in natural language processing. Recent efforts have focused on generating paraphrases in specific syntactic forms, which, generally, heavily relies on manually annotated paraphrase data that is not easily available for many languages and domains. In this paper, we propose a novel end-to-end framework to leverage existing large-scale bilingual parallel corpora to generate paraphrases under the control of syntactic exemplars. In order to train one model over the two languages of parallel corpora, we embed sentences of them into the same content and style spaces with shared content and style encoders using cross-lingual word embeddings. We propose an adversarial discriminator to disentangle the content and style space, and employ a latent variable to model the syntactic style of a given exemplar in order to guide the two decoders for generation. Additionally, we introduce cycle and masking learning schemes to efficiently train the model. Experiments and analyses demonstrate that the proposed model trained only on bilingual parallel data is capable of generating diverse paraphrases with desirable syntactic styles. Fine-tuning the trained model on a small paraphrase corpus makes it substantially outperform state-of-the-art paraphrase generation models trained on a larger paraphrase dataset.

We propose a machine learning based approach to design few-mode DRAs by using neural networks to optimize the pump wavelengths, powers and mode content in order to obtain flat gain spectrum with low mode-dependent gain (MDG). Based on the proposed intelligent inverse design method, amplification optimization for the random fiber laser based two-mode DRA can be achieved with gain flatness of 1.0 dB and MDG of 0.6 dB at 14.5 dB on-off gain level. For backward pumping four-mode DRA, gain flatness of 0.46 dB and MDG of 0.3 dB can be achieved at 12.5 dB on-off gain.

Graphical abstract, Highlights • Large datasets for tooth-marked tongue recognition are constructed. • The deep CNN in artificial intelligence gain classification accuracy over 90%. • The models can be successfully generalized to images with different illuminations. • Isolation of the tongue region can enhance tongue diagnosis performance., Tongue diagnosis plays a pivotal role in traditional Chinese medicine (TCM) for thousands of years. As one of the most important tongue characteristics, tooth-marked tongue is related to spleen deficiency and can greatly contribute to the symptoms differentiation and treatment selection. Yet, the tooth-marked tongue recognition for TCM practitioners is subjective and challenging. Most of the previous studies have concentrated on subjectively selected features of the tooth-marked region and gained accuracy under 80%. In the present study, we proposed an artificial intelligence framework using deep convolutional neural network (CNN) for the recognition of tooth-marked tongue. First, we constructed relatively large datasets with 1548 tongue images captured by different equipments. Then, we used ResNet34 CNN architecture to extract features and perform classifications. The overall accuracy of the models was over 90%. Interestingly, the models can be successfully generalized to images captured by other devices with different illuminations. The good effectiveness and generalization of our framework may provide objective and convenient computer-aided tongue diagnostic method on tracking disease progression and evaluating pharmacological effect from a informatics perspective.

This paper presents the systems developed by Beijing Jiaotong University for the CCMT 2020 quality estimation task. In this paper, we propose an effective method to utilize pretrained language models to improve the performance of QE. Our model combines three popular pretrained models, which are Bert, XLM and XLM-R, to create a very strong baseline for both sentence-level and word-level QE. We tried different strategies, including further pretraining for bilingual input, multi-task learning for multi-granularities and weighted loss for unbalanced word labels. To generate more accurate prediction, we performed model ensemble for both granularities. Experiment results show high accuracy on both directions, and outperform the winning system of last year on sentence level, demonstrating the effectiveness of our proposed method.

Machine translation quality estimation (Quality Estimation, QE) aims to evaluate the quality of machine translation automatically without golden reference. QE is an important component in making machine translation useful in real-world applications. Existing approaches require large amounts of expert annotated data. Recently, there are some trials to perform QE in an unsupervised manner, but these methods are based on glass-box features which demands probation inside the machine translation system. In this paper, we propose a new paradigm to perform unsupervised QE in black-box setting, without relying on human-annotated data or model-related features. We create pseudo-data based on Machine Translation Evaluation (MTE) metrics from existing machine translation parallel dataset, and the data are used to fine-tune multilingual pre-trained language models to fit human evaluation. Experiment results show that our model surpasses the previous unsupervised methods by a large margin, and achieve state-of-the-art results on MLQE Dataset.

This paper presents the systems developed by Beijing Jiaotong University for the CCMT 2020 multilingual translation evaluation task. For this translation task, we need to build a Japanese-English translation system based on only Japanese-Chinese and English-Chinese data. Our method mainly relies on synthetic data generated by back translation. We implemented three different architectures, namely Transformer-big, Transformer-base and Dynamic-Conv. We also implemented multi-model ensemble technique to further boost the final result. Experiments show that our machine translation system achieved high accuracy without relying on any bilingual training data.

The aspect-based sentiment analysis (ABSA) task remains to be a long-standing challenge, which aims to extract the aspect term and then identify its sentiment orientation.In previous approaches, the explicit syntactic structure of a sentence, which reflects the syntax properties of natural language and hence is intuitively crucial for aspect term extraction and sentiment recognition, is typically neglected or insufficiently modeled. In this paper, we thus propose a novel dependency syntactic knowledge augmented interactive architecture with multi-task learning for end-to-end ABSA. This model is capable of fully exploiting the syntactic knowledge (dependency relations and types) by leveraging a well-designed Dependency Relation Embedded Graph Convolutional Network (DreGcn). Additionally, we design a simple yet effective message-passing mechanism to ensure that our model learns from multiple related tasks in a multi-task learning framework. Extensive experimental results on three benchmark datasets demonstrate the effectiveness of our approach, which significantly outperforms existing state-of-the-art methods. Besides, we achieve further improvements by using BERT as an additional feature extractor., Comment: Code: https://github.com/XL2248/DREGCN

In Chinese dependency parsing, the joint model of word segmentation, POS tagging and dependency parsing has become the mainstream framework because it can eliminate error propagation and share knowledge, where the transition-based model with feature templates maintains the best performance. Recently, the graph-based joint model [19] on word segmentation and dependency parsing has achieved better performance, demonstrating the advantages of the graph-based models. However, this work can not provide POS information for downstream tasks, and the POS tagging task was proved to be helpful to the dependency parsing according to the research of the transition-based model. Therefore, we propose a graph-based joint model for Chinese word segmentation, POS tagging and dependency parsing. We designed a character-level POS tagging task, and then train it jointly with the model of [19]. We adopt two methods of joint POS tagging task, one is by sharing parameters, the other is by using tag attention mechanism, which enables the three tasks to better share intermediate information and improve each other’s performance. The experimental results on the Penn Chinese treebank (CTB5) show that our proposed joint model improved by 0.38% on dependency parsing than the model of [19]. Compared with the best transition-based joint model, our model improved by 0.18%, 0.35% and 5.99% respectively in terms of word segmentation, POS tagging and dependency parsing.

Machine translation quality estimation (Quality Estimation, QE) aims to evaluate the quality of machine translation automatically without golden reference. QE can be implemented on different granularities, thus to give an estimation for different aspects of machines translation output. In this paper, we propose an effective method to utilize pretrained language models to improve the performance of QE. Our model combines two popular pretrained models, which are Bert and XLM, to create a very strong baseline for both sentence-level and word-level QE. We also propose a simple yet effective strategy, ensemble distillation, to further improve the accuracy of QE system. Ensemble distillation can integrate different knowledge from multiple models into one model, and strengthen each single model by a large margin. We evaluate our system on CCMT2019 Chinese-English and English-Chinese QE dataset, which contains word-level and sentence-level subtasks. Experiment results show our model surpasses previous models to a large extend, demonstrating the effectiveness of our proposed method.

Paraphrase generation (PG) is of great importance to many downstream tasks in natural language processing. Diversity is an essential nature to PG for enhancing generalization capability and robustness of downstream applications. Recently, neural sequence-to-sequence (Seq2Seq) models have shown promising results in PG. However, traditional model training for PG focuses on optimizing model prediction against single reference and employs cross-entropy loss, which objective is unable to encourage model to generate diverse paraphrases. In this work, we present a novel approach with multi-objective learning to PG. We propose a learning-exploring method to generate sentences as learning objectives from the learned data distribution, and employ reinforcement learning to combine these new learning objectives for model training. We first design a sample-based algorithm to explore diverse sentences. Then we introduce several reward functions to evaluate the sampled sentences as learning signals in terms of expressive diversity and semantic fidelity, aiming to generate diverse and high-quality paraphrases. To effectively optimize model performance satisfying different evaluating aspects, we use a GradNorm-based algorithm that automatically balances these training objectives. Experiments and analyses on Quora and Twitter datasets demonstrate that our proposed method not only gains a significant increase in diversity but also improves generation quality over several state-of-the-art baselines.

We report an intelligent gain flattering method for rapid, precise and objective-driven FMF Raman amplifier design, by using machine learning based inverse design method to optimize the pump wavelengths, powers and mode contents.

More massive volume of data are generated in many areas than ever before. However, the missing of some values in collected data always occurs in practice and challenges extracting maximal value from these large scale data sets. Nevertheless, in multivariable time series, most of the existing methods either might be infeasible or could be inefficient to predict the missing data. In this paper, we have taken up the challenge of missing data prediction in multivariable time series by employing improved matrix factorization techniques. Our approaches are optimally designed to largely utilize both the internal patterns of each time series and the information of time series across multiple sources. Based on the idea, we have imposed three different regularization terms to constrain the objective functions of matrix factorization and built five corresponding models. Extensive experiments on real-world data sets and synthetic data set demonstrate that the proposed approaches can effectively improve the performance of missing data prediction in multivariable time series. Furthermore, we have also demonstrated how to take advantage of the high processing power of Apache Spark to perform missing data prediction in large scale multivariable time series.