A5017835517- Topic Modeling
- Natural Language Processing Techniques
- Adversarial Robustness in Machine Learning
- Advanced Text Analysis Techniques
- Semantic Web and Ontologies
- Access Control and Trust
- Speech Recognition and Synthesis
- Service-Oriented Architecture and Web Services
- Speech and dialogue systems
- Computational and Text Analysis Methods
- Advanced Graph Neural Networks
- Advanced Database Systems and Queries
- Complex Network Analysis Techniques
- Neural Networks and Applications
- Cloud Data Security Solutions
- Text and Document Classification Technologies
- Multimodal Machine Learning Applications
- Advanced Malware Detection Techniques
- Advanced Memory and Neural Computing
- Privacy-Preserving Technologies in Data
- Distributed and Parallel Computing Systems
- Ferroelectric and Negative Capacitance Devices
- Recommender Systems and Techniques
- Text Readability and Simplification
- Multi-Agent Systems and Negotiation
Guilin University of Technology
2024
Fudan University
2015-2024
Tianjin Medical University
2024
Third Affiliated Hospital of Guangzhou Medical University
2024
Guangzhou Medical University
2024
Inspur (China)
2024
Civil Aviation Flight University of China
2023
Xiamen University
2018
General Electric (United States)
2013
Jilin Vocational College of Industry and Technology
2009-2012
This study explores the feasibility of performing Chinese word segmentation (CWS) and POS tagging by deep learning. We try to avoid task-specific feature engineering, use layers neural networks discover relevant features tasks. leverage large-scale unlabeled data improve internal representation characters, these improved representations enhance supervised models. Our achieved close state-of-theart performance with minimal computational cost. also describe a perceptron-style algorithm for...
It is desirable for dialog systems to have capability express specific emotions during a conversation, which has direct, quantifiable impact on improvement of their usability and user satisfaction. After careful investigation real-life conversation data, we found that there are at least two ways with language. One describe emotional states by explicitly using strong words; another increase the intensity experiences implicitly combining neutral words in distinct ways. We propose an dialogue...
Xiao Wang, Qin Liu, Tao Gui, Qi Zhang, Yicheng Zou, Xin Zhou, Jiacheng Ye, Yongxin Rui Zheng, Zexiong Pang, Qinzhuo Wu, Zhengyan Li, Chong Ruotian Ma, Zichu Fei, Ruijian Cai, Jun Zhao, Xingwu Hu, Zhiheng Yan, Yiding Tan, Yuan Qiyuan Bian, Zhihua Shan Qin, Bolin Zhu, Xiaoyu Xing, Jinlan Fu, Yue Minlong Peng, Xiaoqing Yaqian Zhongyu Wei, Xipeng Qiu, Xuanjing Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on...
Yi Zhou, Xiaoqing Zheng, Cho-Jui Hsieh, Kai-Wei Chang, Xuanjing Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.
Recent studies have shown that deep neural network-based models are vulnerable to intentionally crafted adversarial examples, and various methods been proposed defend against word-substitution attacks for NLP models. However, there is a lack of systematic study on comparing different defense approaches under the same attacking setting. In this paper, we seek fill gap through comprehensive researches understanding behavior text classifiers trained by representative attacks. addition, propose...
Spiking neural networks (SNNs) are bio-inspired that model how neurons in the brain communicate through discrete spikes, which have great potential various tasks due to their energy efficiency and temporal processing capabilities. SNNs with self-attention mechanisms (Spiking Transformers) recently shown advancements such as sequential modeling image classifications. However, integrating positional information, is essential for capturing relationships data, remains a challenge Transformers....
Despite achieving prominent performance on many important tasks, it has been reported that neural networks are vulnerable to adversarial examples. Previously studies along this line mainly focused semantic tasks such as sentiment analysis, question answering and reading comprehension. In study, we show examples also exist in dependency parsing: propose two approaches study where how parsers make mistakes by searching over perturbations existing texts at sentence phrase levels, design...
Despite neural networks have achieved prominent performance on many natural language processing (NLP) tasks, they are vulnerable to adversarial examples. In this paper, we propose Dirichlet Neighborhood Ensemble (DNE), a randomized smoothing method for training robust model defense substitution-based attacks. During training, DNE forms virtual sentences by sampling embedding vectors each word in an input sentence from convex hull spanned the and its synonyms, it augments them with data. such...
Existing studies have demonstrated that adversarial examples can be directly attributed to the presence of non-robust features, which are highly predictive, but easily manipulated by adversaries fool NLP models. In this study, we explore feasibility capturing task-specific robust while eliminating ones using information bottleneck theory. Through extensive experiments, show models trained with our bottleneck-based method able achieve a significant improvement in accuracy, exceeding...
Chong Li, Cenyuan Zhang, Xiaoqing Zheng, Xuanjing Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 2021.
A phase retrieval method based on deep learning with bandpass filtering in holographic data storage is proposed. The relationship between the known encoded pages and their near-field diffraction intensity patterns established by an end-to-end convolutional neural network, which used to predict unknown page. We found training efficiency of mainly determined edge details adjacent codes, are high-frequency components code. Therefore, we can attenuate low-frequency reduce material consumption....
Most of today's web content is designed for human consumption, which makes it difficult software tools to access them readily. Even that automatically generated from back-end databases usually presented without the original structural information. In this paper, we present an automated information extraction algorithm can extract relevant attribute-value pairs product descriptions across different sites. A notion, called structural-semantic entropy, used locate data interest on pages,...
A sequence-to-sequence (seq2seq) learning with neural networks empirically shows to be an effective framework for grammatical error correction (GEC), which takes a sentence errors as input and outputs the corrected one. However, performance of GEC models seq2seq heavily relies on size quality corpus hand. We propose method inspired by adversarial training generate more meaningful valuable examples continually identifying weak spots model, enhance model gradually adding generated set....
Unsupervised word representations have demonstrated improvements in predictive generalization on various NLP tasks. Most of the existing models are fact good at capturing relatedness among words rather than their ''genuine'' similarity because context often represented by a sum (or an average) neighbor's embeddings, which simplifies computation but ignores important that meaning is determined its context, reflecting not only surrounding also rules used to combine them (i.e....
We propose a convolutional neural network architecture with k-max pooling layer for semantic modeling of music. The aim music model is to analyze and represent the content purposes classification, discovery, or clustering. used in make it possible pool k most active features, capturing semantic-rich time-varying information about Our takes an input as sequence audio words, where each word associated distributed feature vector that can be fine-tuned by backpropagating errors during training....
Deep neural networks are vulnerable to adversarial attacks, where a small perturbation an input alters the model prediction. In many cases, malicious inputs intentionally crafted for one can fool another model. this paper, we present first study systematically investigate transferability of examples text classification models and explore how various factors, including network architecture, tokenization scheme, word embedding, capacity, affect examples. Based on these studies, propose genetic...
In the context of carbon neutrality, low-carbon transition in agriculture is crucial to achieving mitigation through clean production. The provision agricultural productive services (APS) pivotal for modernizing farming practices China. However, impact this on transformation has received limited attention. This research examined non-linear relationship between and development, including verifying a threshold effect with APS as variable, employing panel data 31 provinces China from 2010 2021....
In the process of zinc hydrometallurgy, content fluorine in sulfate solution directly affects stripping plate, which easily leads to deterioration working conditions. It not only has a serious impact on entire hydrometallurgical system but also causes huge economic losses. Especially secondary resource utilization, concentration fluoride ions electrolyte exceeds control standard smelting enterprises, become long-term technical challenge industry. So far, no efficient and economical been...
Very recently, some studies on neural dependency parsers have shown advantage over the traditional ones a wide variety of languages. However, for graph-based parsing systems, they either count long-term memory and attention mechanism to implicitly capture high-order features or give up global exhaustive inference algorithms in order harness rich history decisions. The former might miss out important specific headword predictions without help explicit structural information, latter may suffer...