A5101849569- Topic Modeling
- Natural Language Processing Techniques
- Multimodal Machine Learning Applications
- Advanced Text Analysis Techniques
- Text and Document Classification Technologies
- Quantum Chromodynamics and Particle Interactions
- Quantum Dots Synthesis And Properties
- Advanced Graph Neural Networks
- Web Data Mining and Analysis
- Parallel Computing and Optimization Techniques
- Chalcogenide Semiconductor Thin Films
- Fuzzy Systems and Optimization
- Copper-based nanomaterials and applications
- Advanced Image and Video Retrieval Techniques
- Biomedical Text Mining and Ontologies
- Particle physics theoretical and experimental studies
- Information Retrieval and Search Behavior
- Expert finding and Q&A systems
- Radiomics and Machine Learning in Medical Imaging
- Cancer-related molecular mechanisms research
- Sentiment Analysis and Opinion Mining
- ICT Impact and Policies
- Data Quality and Management
- Hate Speech and Cyberbullying Detection
- Access Control and Trust
State Key Laboratory of Cryptology
2024
Chengdu University of Information Technology
2023-2024
China People's Public Security University
2023
Chongqing University of Posts and Telecommunications
2013-2022
Guizhou University
2022
Zhejiang University
2022
Tiangong University
2022
Microsoft Research (United Kingdom)
2020-2021
University of California, Los Angeles
2021
Microsoft (United States)
2020
This paper presents a new sequence-to-sequence pre-training model called ProphetNet, which introduces novel self-supervised objective named future n-gram prediction and the proposed n-stream self-attention mechanism. Instead of optimizing one-step-ahead in traditional model, ProphetNet is optimized by n-step ahead that predicts next n tokens simultaneously based on previous context at each time step. The explicitly encourages to plan for prevent overfitting strong local correlations. We...
Abstract To emphasize the semantic impact of local and grammatical information among adjacent words in input text, we establish a constraint functions-based quantum-like tensor compression sentence representation model by integrating concept extending pure state-based density matrix to mixed-state projection operator quantum mechanics. The provided highlights significance mixed word associations simultaneously reducing reliance on derived solely from dictionary statistics. We combine...
Based on the fact that Hukuhara difference exists only under very restrictive conditions, in this paper, we present process of computing generalized discrete Z-numbers and continuous respectively. Some examples are given to illustrate effectiveness proposed methods.
Reading long documents to answer open-domain questions remains challenging in natural language understanding. In this paper, we introduce a new model, called RikiNet, which reads Wikipedia pages for question answering. RikiNet contains dynamic paragraph dual-attention reader and multi-level cascaded predictor. The dynamically represents the document by utilizing set of complementary attention mechanisms. representations are then fed into predictor obtain span short answer, type manner. On...
This paper presents a new sequence-to-sequence pre-training model called ProphetNet, which introduces novel self-supervised objective named future n-gram prediction and the proposed n-stream self-attention mechanism. Instead of optimizing one-step-ahead in traditional model, ProphetNet is optimized by n-step ahead that predicts next n tokens simultaneously based on previous context at each time step. The explicitly encourages to plan for prevent overfitting strong local correlations. We...
There has been a steady need in the medical community to precisely extract temporal relations between clinical events. In particular, information can facilitate variety of downstream applications such as case report retrieval and question answering. Existing methods either require expensive feature engineering or are incapable modeling global relational dependencies among this paper, we propose novel method, Clinical Temporal ReLation Exaction with Probabilistic Soft Logic Regularization...
Dayiheng Liu, Yu Yan, Yeyun Gong, Weizhen Qi, Hang Zhang, Jian Jiao, Weizhu Chen, Jie Fu, Linjun Shou, Ming Pengcheng Wang, Jiusheng Daxin Jiang, Jiancheng Lv, Ruofei Winnie Wu, Zhou, Nan Duan. Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021.
In this paper, we propose a novel data augmentation method, referred to as Controllable Rewriting based Question Data Augmentation (CRQDA), for machine reading comprehension (MRC), question generation, and question-answering natural language inference tasks. We treat the task constrained rewriting problem generate context-relevant, high-quality, diverse samples. CRQDA utilizes Transformer Autoencoder map original discrete into continuous embedding space. It then uses pre-trained MRC model...
Weizhen Qi, Yeyun Gong, Yu Yan, Can Xu, Bolun Yao, Bartuer Zhou, Biao Cheng, Daxin Jiang, Jiusheng Chen, Ruofei Zhang, Houqiang Li, Nan Duan. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing: System Demonstrations. 2021.
Abstract Natural answer generation is in a very clear practical significance and strong application background, which can be widely used the field of knowledge services such as community question answering intelligent customer service. Traditional to provide precise entities neglect defects; namely, users hope receive complete natural answer. In this research, we propose novel attention-based recurrent neural network for generation, enhanced with multi-level copying mechanisms question-aware...
We present a systematic study of the production heavy quarkonium, i.e., $|(c\bar{c})[n] \rangle$ , $|(b\bar{c})[n] (or $|(c\bar{b})[n] \rangle$), and $|(b\bar{b})[n] quarkonium [$|(Q\bar{Q'})[n]\rangle$ for short], through $Z^0$ boson semi-exclusive decays with new parameters \cite{lx} under framework NRQCD, where $[n]$ stands $n^1S_0$, $n^3S_1$, $n^1P_0$, $n^3P_J$ ($n=1, \cdots, 6$; $J=(0, 1, 2)$). "Improved trace technology" is adopted to derive simplified analytic expressions at amplitude...
News headline generation aims to produce a short sentence attract readers read the news. One news article often contains multiple keyphrases that are of interest different users, which can naturally have reasonable headlines. However, most existing methods focus on single generation. In this paper, we propose generating headlines with user interests, whose main idea is generate users for first, and then keyphrase-relevant We multi-source Transformer decoder, takes three sources as inputs:...
The production of the heavy quarkonium, i.e., $|(c\overline{b})[n]⟩$ (or $|(b\overline{c})[n]⟩$), $|(c\overline{c})[n]⟩$, and $|(b\overline{b})[n]⟩$- quarkonium [$|(Q\overline{{Q}^{\ensuremath{'}}})[n]⟩$-quarkonium for short], through Higgs ${H}^{0}$ boson semiexclusive decays is evaluated within nonrelativistic quantum chromodynamics (NRQCD) framework, where [$n$] stands two color-singlet $S$-wave states, $|(Q\overline{{Q}^{\ensuremath{'}}})[{^{1}S}_{0}{]}_{\mathbf{1}}⟩$...
With the formation and popularity of Internet Things(IoT), difficulty protecting IoT infrastructure smart devices from a few-shot ever-changing malicious attacks has increased significantly. Traditional intrusion detection models in static mode cannot defend against intelligent that change real time are good at reconnaissance, it is difficult to achieve effective attacks. Therefore, solve above problems, this paper proposes variable model GDE Model for IoT, which contains data processing...
Multi-task benchmarks such as GLUE and SuperGLUE have driven great progress of pretraining transfer learning in Natural Language Processing (NLP). These mostly focus on a range Understanding (NLU) tasks, without considering the Generation (NLG) models. In this paper, we present General Evaluation (GLGE), new multi-task benchmark for evaluating generalization capabilities NLG models across eight language generation tasks. For each task, continue to design three subtasks terms task difficulty...
Reading long documents to answer open-domain questions remains challenging in natural language understanding. In this paper, we introduce a new model, called RikiNet, which reads Wikipedia pages for question answering. RikiNet contains dynamic paragraph dual-attention reader and multi-level cascaded predictor. The dynamically represents the document by utilizing set of complementary attention mechanisms. representations are then fed into predictor obtain span short answer, type manner. On...
Yu Yan, Fei Hu, Jiusheng Chen, Nikhil Bhendawade, Ting Ye, Yeyun Gong, Nan Duan, Desheng Cui, Bingyu Chi, Ruofei Zhang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing: System Demonstrations. 2021.
It is important to learn directly from original texts in natural language processing (NLP). Many deep learning (DP) models needing a large number of manually annotated data are not effective deriving much information corpora with few labels. Existing methods using unlabeled provide valuable messages consume considerable time and cost. Our provided sentence representation based on quantum computation (called Model I) needs no prior knowledge except word2vec. To reduce some semantic noise...
News headline generation aims to produce a short sentence attract readers read the news. One news article often contains multiple keyphrases that are of interest different users, which can naturally have reasonable headlines. However, most existing methods focus on single generation. In this paper, we propose generating headlines with user interests, whose main idea is generate users for first, and then keyphrase-relevant We multi-source Transformer decoder, takes three sources as inputs:...
Now, the pre-training technique is ubiquitous in natural language processing field. ProphetNet a based generation method which shows powerful performance on English text summarization and question tasks. In this paper, we extend into other domains languages, present family models, named ProphetNet-X, where X can be English, Chinese, Multi-lingual, so on. We pre-train cross-lingual model ProphetNet-Multi, Chinese ProphetNet-Zh, two open-domain dialog models ProphetNet-Dialog-En...