Zhen Xu

ORCID: 0000-0001-9688-6958
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Topic Modeling
  • Advanced Image and Video Retrieval Techniques
  • Speech and dialogue systems
  • Natural Language Processing Techniques
  • Image Retrieval and Classification Techniques
  • Music and Audio Processing
  • Advanced Computational Techniques and Applications
  • Web Data Mining and Analysis
  • Machine Learning and ELM
  • Spectroscopy and Chemometric Analyses
  • Model Reduction and Neural Networks
  • Advanced Neural Network Applications
  • Video Analysis and Summarization
  • Remote-Sensing Image Classification
  • Speech Recognition and Synthesis
  • Handwritten Text Recognition Techniques
  • Sentiment Analysis and Opinion Mining
  • Mineral Processing and Grinding
  • Generative Adversarial Networks and Image Synthesis
  • Water Quality Monitoring and Analysis
  • Multimodal Machine Learning Applications
  • Speech and Audio Processing
  • Text and Document Classification Technologies
  • Domain Adaptation and Few-Shot Learning
  • Economic and Industrial Development

Harbin Institute of Technology
2017-2023

Heilongjiang Provincial Academy of Agricultural Sciences
2016-2022

Nvidia (United Kingdom)
2022

Tencent (China)
2022

Lenovo (China)
2021

Sichuan Normal University
2011-2021

Institute of Information Engineering
2021

Chinese Academy of Sciences
2021

Nanjing Normal University
2021

Google (United States)
2019-2020

This paper presents a Generative Adversarial Network (GAN) to model single-turn short-text conversations, which trains sequence-to-sequence (Seq2Seq) network for response generation simultaneously with discriminative classifier that measures the differences between human-produced responses and machine-generated ones. In addition, proposed method introduces an approximate embedding layer solve non-differentiable problem caused by sampling-based output decoding procedure in Seq2Seq generative...

10.18653/v1/d17-1065 article EN cc-by Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing 2017-01-01

Automatic chatbots (also known as chat-agents) have attracted much attention from both researching and industrial fields. Generally, the semantic relevance between users' queries corresponding responses is considered essential element for conversation modeling in generation ranking based chat systems. By contrast, it a nontrivial task to adopt information, such preference, social role, etc., into conversational models reasonably, while profiles play significant role procedure of...

10.1109/taslp.2017.2763243 article EN IEEE/ACM Transactions on Audio Speech and Language Processing 2017-10-25

It is critical for automatic chat-bots to gain the ability of conversation comprehension, which essence provide context-aware responses conduct smooth dialogues with human beings. As basis this task, modeling will notably benefit from background knowledge, since such knowledge indeed implicates semantic hints that help further clarify relationships between sentences within a conversation. In paper, deep neural network proposed incorporate modeling. Through recall mechanism specially designed...

10.1109/ijcnn.2017.7966297 article EN 2022 International Joint Conference on Neural Networks (IJCNN) 2017-05-01

This paper studies a training method to jointly estimate an energy-based model and flow-based model, in which the two models are iteratively updated based on shared adversarial value function. joint has following traits. (1) The update of is noise contrastive estimation, with flow serving as strong distribution. (2) approximately minimizes Jensen-Shannon divergence between data (3) Unlike generative networks (GAN) estimates implicit probability distribution defined by generator our explicit...

10.1109/cvpr42600.2020.00754 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2020-06-01

Visual Information Extraction (VIE) task aims to extract key information from multifarious document images (e.g., invoices and purchase receipts). Most previous methods treat the VIE simply as a sequence labeling problem or classification problem, which requires models carefully identify each kind of semantics by introducing multimodal features, such font, color, layout. But features can't work well when faced with numeric semantic categories some ambiguous texts. To address this issue, in...

10.24963/ijcai.2021/144 article EN 2021-08-01

A project's documentation is the primary source of information for developers using that project. With hundreds thousands programming-related questions posted on programming Q&A websites, such as Stack Overflow, we question whether developer-written provides enough guidance programmers. In this study, wanted to know if there are any topics which inadequately covered by project documentation. We combined from Overflow and PHP Python projects. Then, applied topic analysis data latent Dirichlet...

10.1109/msr.2013.6624005 article EN 2013-05-01

It's been a long history that most object detection methods obtain objects by using the non-maximum suppression (NMS) and its improved versions like Soft-NMS to remove redundant bounding boxes. We challenge those NMS-based from three aspects: 1) The box with highest confidence value may not be true positive having biggest overlap ground-truth box. 2) Not only is required for boxes, but also enhancement needed positives. 3) Sorting candidate boxes values necessary so full parallelism...

10.1109/cvpr52688.2022.00122 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2022-06-01

Band selection is a well-known approach for reducing dimensionality in hyperspectral imaging. In this paper, band method based on consistency-measure of neighborhood rough set theory (CMNRS) was proposed to select informative bands from images. A decision-making information system established by the reflection spectrum soybeans' data between 400 nm and 1000 wavelengths. The consistency-measure, which reflects not only size decision positive region, but also sample distribution boundary used...

10.1088/0957-0233/27/5/055501 article EN Measurement Science and Technology 2016-04-06

Band selection is considered to be an important processing step in handling hyperspectral data. In this work, we selected informative bands according the maximal relevance minimal redundancy (MRMR) criterion based on neighborhood mutual information. Two measures MRMR difference and quotient were defined a forward greedy search for band was constructed. The performance of proposed algorithm, along with comparison other methods (neighborhood dependency measure genetic algorithm uninformative...

10.1088/0957-0233/27/12/125501 article EN Measurement Science and Technology 2016-11-01

Diarrhetic shellfish poisoning (DSP) toxins are potent marine biotoxins. It can cause a severe gastrointestinal illness by the consumption of mussels contaminated DSP toxins. New methods for effectively and rapidly detecting toxins-contaminated required. In this study, we used near-infrared (NIR) reflection spectroscopy combined with pattern recognition to detect range 950-1700 nm, spectral data healthy were acquired. To select optimal waveband subsets, selection algorithm Gaussian...

10.3389/fmars.2022.907378 article EN cc-by Frontiers in Marine Science 2022-06-09

Zhen Xu, Nan Jiang, Bingquan Liu, Wenge Rong, Bowen Wu, Baoxun Wang, Zhuoran Xiaolong Wang. Proceedings of the 2018 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

10.18653/v1/n18-1188 article EN cc-by 2018-01-01

This article aims to develop a method detect visual differences introduced into web pages when they are rendered in different browsers. To achieve this goal, we propose an empirical similarity metric by mimicking human mechanisms of perception. The Gestalt laws grouping translated computer compatible rule set. A block tree is then parsed the rules for calculation. During translation laws, experiments performed obtain metrics proximity, color similarity, and image similarity. After validation...

10.1145/3140544 article EN ACM Transactions on Internet Technology 2018-04-17

In human-to-human conversations, the context generally provides several backgrounds and strategic points for following response. Therefore, many response generation approaches have explored methodologies to incorporate into encoder-decoder architecture, generate context-aware responses that are remarkably relevant cohesive given context. However, most pay less attention semantic interactions implicitly existing within contextual utterances, which of great importance capture clues dialog...

10.1109/taslp.2019.2915922 article EN IEEE/ACM Transactions on Audio Speech and Language Processing 2019-05-24

In order to better manage and protect rivers lakes, the most important requirement is find objects on surface of lakes in time. Generally, image segmentation target detection are used detect water targets. The former sensitive selection features, with poor generalization ability slow speed. latter has not yet been applied UAV images. view this situation, paper proposes a model based YOLOv3, which targets verify performance model, images collected include five types These then enhanced by...

10.1145/3456415.3456424 article EN 2021-02-25

With the rapid advancements in Augmented Reality, number of AR users is gradually increasing and multiuser ecosystem on rise. Currently, applications usually present results without limitations, which causes great latent danger to users, so it necessary apply strategies ensure safe output AR. Due environmental diversities among distributed traditional approaches designed for single-user are not efficient multi-user applications. Considering characteristics scenarios, we propose a strategy...

10.1109/iscc53001.2021.9631507 article EN 2022 IEEE Symposium on Computers and Communications (ISCC) 2021-09-05

Traditional text-based web page similarity measures fail to handle rich-information-embedded modern pages. Current approaches regard pages as either DOM trees or images. However, the former only focuses on structure, while latter ignores inner connections among different features. Therefore, they are not suitable for Hence, idea of a block tree is introduced, which contains both structural and visual information A metric proposed edit distance between two trees. Finally, an experiment...

10.1504/ijwet.2017.086415 article EN International Journal of Web Engineering and Technology 2017-01-01

The Cognitive Radio (CR) technology is an efficient solution to spectrum scarcity by share the with secondary users on a non-interfering basis. prediction can rationalize allocation based previous information about evolution in time. Against algorithm lack of timeliness and accuracy, this paper proposes novel approach for Optimally Pruned Extreme Learning Machine (OP-ELM) which improved original (ELM) algorithm. This method not only takes advantage ELM extremely fast speed good precision,...

10.4028/www.scientific.net/amm.536-537.430 article EN Applied Mechanics and Materials 2014-04-01
Coming Soon ...