Xiong Yu-Jie

Xiong Yu-Jie

Associate professor

Sitemap

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

🎉 Our Team’s Paper Accepted to ICCV 2025! 🎉

less than 1 minute read

Published:

We are thrilled to announce that our team’s paper, “MSA: Multi-view Character Representation for Open-set Chinese Text Recognition”, has been officially accepted to ICCV 2025 — one of the top-tier conferences in computer vision!

Our paper has been accepted to EMNLP 2025!

less than 1 minute read

Published:

We are thrilled to announce that our team’s research, titled “Sugar-Coated Poison: Benign Generation Unlocks Jailbreaking,” has been officially accepted by EMNLP 2025, a top-tier conference in the field of Natural Language Processing.

portfolio

publications

71.Text-independent Writer Identification Using SIFT Descriptor and Contour-directional Feature

Published in International Conference on Document Analysis and Recognition, 2015

A method using SIFT and CDF for text - independent writer identification is proposed. It has two - stage processing and outperforms other algorithms on two datasets.

Recommended citation:

Text-independent Writer Identification Using SIFT Descriptor and Contour-directional Feature, Y.-J. Xiong, Y. Wen, Patrick. S. P. Wang and Y. Lu*, in Proceedings of the International Conference on Document Analysis and Recognition, 2015, 91–95

Download Paper

70.基于两方向动态时间规整的无分割手写汉字检测

Published in 计算机应用研究, 2016

The paper proposes a method combining SIFT keypoint location and two - directional DTW for Chinese handwritten character detection without segmentation. It shows good results but has limitations.

Recommended citation:

基于两方向动态时间规整的无分割手写汉字检测, 黄志敏*，姚舜奕，熊玉洁, 《计算机应用研究》，2016，33.11: 3499–3502

Download Paper

69.Off-line Text-Independent Writer Recognition: A Survey

Published in International Journal of Pattern Recognition and Artificial Intelligence, 2017

This paper focuses on off - line text - independent writer recognition. It summarizes methods, shows datasets, and compares performances. Spatial features outperform others in some aspects.

Recommended citation:

Off-line Text-independent Writer Recognition: A Survey, Y.-J. Xiong, Y. Lu* and Patrick. S. P. Wang, International Journal of Pattern Recognition and Artificial Intelligence, 2017, 31.5: 1756008

Download Paper

68.Off-line Text-independent Writer Identification for Chinese Handwriting

Published in Series on Language Processing, Pattern Recognition, and Intelligent Systems, 2017

A method using CDF and modified SIFT for Chinese writer identification is proposed. It outperforms others, achieving high accuracy on HIT - MW dataset.

Recommended citation:

Off-line Text-independent Writer Identification for Chinese Handwriting, Y.-J. Xiong and Y. Lu*, Advances in Chinese document and text processing, Series on Language Processing, Pattern Recognition, and Intelligent Systems, 2017, 2.8: 215-234

Download Paper

67.Chinese Writer Identification Using Contour-Directional Feature and Character Pair Similarity Measurement

Published in International Conference on Document Analysis and Recognition, 2017

A method using CDF and CPSM for Chinese writer identification fuses two similarities. It outperforms previous methods with high Top - 1 accuracy on two datasets.

Recommended citation:

Chinese Writer Identification Using Contour-Directional Feature and Character Pair Similarity Measurement, Y.-J. Xiong and Y. Lu*, in Proceedings of the International Conference on Document Analysis and Recognition, 2017, 119–124

Download Paper

66.Improving Text-Independent Chinese Writer Identification with the Aid of Character Pairs

Published in International Journal of Pattern Recognition and Artificial Intelligence, 2019

This paper improves text - independent Chinese writer identification by using the similarity of character pairs. It proposes ECF - based scheme and DFS, and re - ranks candidates. Evaluated on two datasets, it outperforms existing methods with high Top - 1 accuracy.

Recommended citation:

Improving Text-Independent Chinese Writer Identification with the Aid of Character Pairs, Y.-J. Xiong, L. Liu, S.-J. Lyu, Patrick S. P. Wang and Y. Lu*, International Journal of Pattern Recognition and Artificial Intelligence, 2019, 33.2: 1953001

Download Paper

65.Improving Chinese Writer Identification by Fusion of Text-dependent and Text-independent Methods

Published in Frontiers in Patten Recognition and Artificial Intelligence, 2019

A method for Chinese writer identification uses text - independent and text - dependent features. It fuses two similarities. Experiments show it outperforms previous methods with high Top - 1 accuracy.

Recommended citation:

Improving Chinese Writer Identification by Fusion of Text-dependent and Text-independent Methods, Y.-J. Xiong, L. Liu, Patrick S. P. Wang and Y. Lu*, Frontiers in Patten Recognition and Artificial Intelligence, Series on Language Processing, Pattern Recognition, and Intelligent Systems, 2019, 5.6: 97-112

Download Paper

64.A Lightweight Improved U-Net with Shallow Features Combination and Its Application to Defect Detection

Published in Wuhan University Journal of Natural Sciences, 2020

A lightweight IU - Net with shallow features combination is proposed, reducing drawbacks, and is applied to detect small metal product defects, outperforming other methods.

Recommended citation:

A Lightweight Improved U-Net with Shallow Features Combination and Its Application to Defect Detection H. Wu, X.-K. Sun*, Y.-J. Xiong, Wuhan University Journal of Natural Sciences, 2020, 25.5: 461-468

Download Paper

63.Handwriting and Hand-Sketched Graphics Detection Using Convolutional Neural Networks

Published in International Conference on Pattern Recognition and Artificial Intelligence, 2020

The paper presents two CNN - based methods, one using CTPN for handwriting detection and the other using Mask - RCNN for hand - sketched graphics detection, and validates their effectiveness on the SUES - 1000 database.

Recommended citation:

Handwriting and Hand-Sketched Graphics Detection Using Convolutional Neural Networks, S.-Y. Cheng, Y.-J. Xiong*, J.-Q. Zhang and Y.-C. Cao, in Proceedings of the International Conference on Pattern Recognition and Artificial Intelligence, 2020, 352-362

Download Paper

62.改进遗传算法的无人机路径规划

Published in 导航定位学报, 2020

The paper proposes an improved genetic algorithm - based path planning method for UAVs to ensure flight safety and shorter distances, and verifies its superiority through experiments.

Recommended citation:

改进遗传算法的无人机路径规划, 吕倩, 孙宪坤*, 熊玉洁, 《导航定位学报》，2020, 8.5: 42-48

Download Paper

61.基于改进ADRC的四旋翼无人机抗干扰姿态控制系统设计

Published in 光电与控制, 2020

An improved ADRC - based attitude control system for quadrotor UAVs, combining GFTSM, is designed and its superiority is verified by simulations.

Recommended citation:

基于改进ADRC的四旋翼无人机抗干扰姿态控制系统设计, 余小燕, 孙宪坤*, 熊玉洁, 胡清礼, 陈善鹏, 《电光与控制》, 2020, 27.12: 78-83

Download Paper

60.Attention U-Net with Multilevel Fusion for License Plate Detection

Published in Wuhan University Journal of Natural Sciences, 2021

The paper presents an AUMF for license plate detection. It details its architecture, loss function, validates on AOLP dataset, and shows better performance in complex conditions.

Recommended citation:

Attention U-Net with Multilevel Fusion for License Plate Detection, Y. Yao, Y.-J. Xiong*, B. Huang and J. Yang, Wuhan University Journal of Natural Sciences, 2021, 26.3: 227-234

Download Paper

59.An Empirical Study of Text Factors and Their Effects on Chinese Writer Identification

Published in Digital TV and Wireless Multimedia Communication, 2021

This paper empirically examines the effects of text factors on Chinese writer identification with text - independent features. It concludes that more characters boost performance, 50 is the minimum number needed, and the number of same characters has little impact above 50.

Recommended citation:

An Empirical Study of Text Factors and Their Effects on Chinese Writer Identification, Y.-J. Xiong*, Y. Lu and Y.-C. Cao, Digital TV and Wireless Multimedia Communication, 2021, 194-205

Download Paper

58.结合倒置特征金字塔和U-Net的高光谱图像分类

Published in 中国图象图形学报, 2021

This paper proposes a method combining inverted feature pyramid and U - Net for hyperspectral image classification. It uses PCA for preprocessing, and experiments show high accuracy and analyze related factors.

Recommended citation:

结合倒置特征金字塔和U-Net的高光谱图像分类, 程嵩阳, 熊玉洁*，姚瑶, 李庆利, 《中国图象图形学报》，2021, 26.8: 1994-2008

Download Paper

57.Attention U-Net with Feature Fusion Module for Robust Defect Detection

Published in Journal of Circuits, Systems and Computers, 2021

U-Net is good at medical image segmentation but not for industrial defect detection. We propose an attention U-Net with a feature fusion module. It combines features and uses attention gates. Experiments on two datasets show it outperforms other methods and has application potential.

Recommended citation:

Attention U-Net with Feature Fusion Module for Robust Defect Detection, Y.-J. Xiong*, Y.-B. Gao, H. Wu and Y. Yao, Journal of Circuits, Systems and Computers, 2021, 30.15: 2150272

Download Paper

56.PC-SuperPoint: Interest Point Detection and Descriptor Extraction Using Pyramid Convolution and Circle Loss

Published in Journal of Electronic Imaging, 2021

We propose PC-SuperPoint using pyramid convolution and circle loss for interest point tasks. Pyramid convolutions extract multiscale features, circle loss aids training, and experiments on relevant datasets show its effectiveness.

Recommended citation:

PC-SuperPoint: Interest Point Detection and Descriptor Extraction Using Pyramid Convolution and Circle Loss, Y.-J. Xiong*, S. Ma, Y.-B. Gao and Z.-J. Fang, Journal of Electronic Imaging, 2021, 30.3: 033024

Download Paper

55.Attention Based Multiple Siamese Network for Offline Signature Verification

Published in International Conference on Document Analysis and Recognition, 2021

The paper presents an attention - based Multiple Siamese Network for offline signature verification. It uses attention modules and contrastive pairs, and shows better performance than previous methods on multiple datasets.

Recommended citation:

Attention Based Multiple Siamese Network for Offline Signature Verification, Y.-J. Xiong* and S.-Y. Cheng, in Proceedings of the International Conference on Document Analysis and Recognition, 2021, 337-349

Download Paper

54.Generalized Multi-view Learning Based on Generalized Eigenvalues Proximal Support Vector Machines

Published in Expert Systems With Applications, 2022

The paper presents GMGSVMs and GMIGSVMs for generalized multi - view learning, uses an alternating algorithm for optimization, and proves their superiority via experiments.

Recommended citation:

Generalized Multi-view Learning Based on Generalized Eigenvalues Proximal Support Vector Machines, X.-J. Xie* and Y.-J. Xiong, Expert Systems with Applications, 2022, 194.1: 116491

Download Paper

53.Knowledge Distilled Pre-training Model for Vision-language-navigation

Published in Applied Intelligence, 2022

The paper presents a knowledge - distilled pre - training model for VLN. It shrinks model size and inference time, keeps 95% of the original performance, and outperforms baselines.

Recommended citation:

Knowledge Distilled Pre-training Model for Vision-language-navigation, B. Huang*, S. Zhang, J.-T. Huang, Y.-J. Yu, Z.-C. Shi and Y.-J. Xiong, Applied Intelligence, 2022, 53.1: 5607–5619

Download Paper

52.基于PWC-Net的多层权值和轻量化改进光流估计算法

Published in 计算机应用研究, 2022

The paper presents a lightweight DS - PWC model based on PWC - Net for optical flow estimation. It uses deep - separable convolutions and data enhancement, achieving 58 fps with good quality and validating its effectiveness.

Recommended citation:

基于 PWC-Net 的多层权值和轻量化改进光流估计算法, 胡毅轩，吴飞*，熊玉洁, 《计算机应用研究》，2022, 39.1: 291-295

Download Paper

51.A Cross Entropy Based Approach to Minimum Propagation Latency for Controller Placement in Software Defined Network

Published in Computer Communications, 2022

The paper proposes a cross - entropy approach to solve it, and validates its effectiveness via experiments.

Recommended citation:

A Cross Entropy Based Approach to Minimum Propagation Latency for Controller Placement in Software Defined Network, J. Chen, Y.-J. Xiong*, X.-H. Qiu, D. He, H.-M. Yin and Y.-F. Xiao, Computer Communications, 2022, 191.1: 133-144

Download Paper

50.License Plate Detection with Attention-Guided Dual Feature Pyramid Networks in Complex Environments

Published in Electronics, 2022

The paper proposes an attention - guided dual feature pyramid network - based license plate detection method for complex environments, validates it on datasets, and proves its superiority.

Recommended citation:

License Plate Detection with Attention-Guided Dual Feature Pyramid Networks in Complex Environments, Y.-J. Xiong*, Y.-B. Gao*, J.-Q. Zhang and J.-X. Ren, Electronics, 2022, 11.23: 3895

Download Paper

49.A Density-based Controller Placement Algorithm for Software Defined Networks

Published in International Conference on Cyber, Physical and Social Computing, 2022

The paper presents an improved DCPA for CPP. It calculates controller numbers, uses multiple indicators and K - means, and experiments show it gets low - cost solutions close to the optimal.

Recommended citation:

A Density-based Controller Placement Algorithm for Software Defined Networks, J. Chen, Y.-J. Xiong* and D. He, in Proceedings of the International Conference on Cyber, Physical and Social Computing, 2022, 287-291

Download Paper

48.SET: A Squeeze-and-excitation Transformer for Offline Signature Verification

Published in International Conference on Ubiquitous Intelligence and Computing, 2022

The paper presents a SET for offline signature verification, using a modified Swin - Transformer and SE module, and it outperforms existing methods on multiple datasets.

Recommended citation:

SET: A Squeeze-and-excitation Transformer for Offline Signature Verification, J.-X. Ren, J. Chen* and Y.-J. Xiong*,in Proceedings of the International Conference on Ubiquitous Intelligence and Computing, 2022, 1812-1816

Download Paper

46.基于Transformer与Vector Loss模块的椎骨Cobb角点定位网络

Published in 中国医学物理学杂志, 2022

This paper presents a vertebral corner detection framework with an embedded Transformer mechanism for calculating Cobb angles. It uses data augmentation, Transformer, and Vector Loss modules to solve automated measurement issues. Experiments on the MICCAI 2019 dataset show the method has high accuracy (SMAPE of 9.01, 1.80 improvement), and can help clinical decision - making. Future work will focus on reducing model depth and complexity.

Recommended citation:

基于Transformer与Vector Loss模块的椎骨Cobb角点定位网络, 陈瑶，高永彬*，熊玉洁, 《中国医学物理学杂志》, 2022，39.11: 1393-1400

Download Paper

45.Mitigating Lifetime-Energy-Makespan Issues in Reliability-Aware Workflow Scheduling for Big Data

Published in Journal of Circuits, Systems, and Computers, 2022

In the big data era, conventional RWS in cloud computing has issues. We propose a new methodology. Simulations show our RWS strategies are superior and the method has potential for big data systems.

Recommended citation:

Mitigating Lifetime-Energy-Makespan Issues in Reliability-Aware Workflow Scheduling for Big Data, Y.-J. Xiong*, S.-Y. Cheng and B. Chen, Journal of Circuits, Systems and Computers, 2022, 31.1: 2250012

Download Paper

44.Learning Transferable Feature Representation with Swin Transformer for Object Recognition

Published in Neural Processing Letters , 2023

Deep learning in computer vision is limited by data - scale dependence. This paper uses Swin Transformer with fine - tuning to overcome data shortage, showing good small - scale dataset object - recognition performance.

Recommended citation:

Learning Transferable Feature Representation with Swin Transformer for Object Recognition, J.-X. Ren, Y.-J. Xiong*, X.-J. Xie and Y.-F. Dai, Neural Processing Letters, 2023, 55.3: 2211–2223

Download Paper

43.结合双金字塔特征融合与级联定位的车牌检测

Published in 计算机工程与应用, 2023

The paper “License Plate Detection Using Siamese Feature Pyramid and Cascaded Positioning” presents an algorithm with a Siamese feature pyramid and cascaded positioning. It performs better than traditional methods on relevant datasets.

Recommended citation:

结合双金字塔特征融合与级联定位的车牌检测, 张俊青，熊玉洁*，孙宪坤，高永彬, 《计算机工程与应用》，2023，59.2: 240-252

Download Paper

42.Attention-based Multiple Siamese Networks with Primary Representation Guiding for Offline Signature Verification

Published in International Journal on Document Analysis and Recognition, 2023

The paper “Attention-based multiple siamese networks with primary representation guiding for offline signature verification” presents a method with siamese networks and special modules. It outperforms others on multiple datasets in offline signature verification.

Recommended citation:

Attention-based Multiple Siamese Networks with Primary Representation Guiding for Offline Signature Verification, Y.-J. Xiong*, S.-Y. Cheng, J.-X. Ren and Y.-J. Zhang, International Journal on Document Analysis and Recognition, 2023， 27： 195–208

Download Paper

41.PDCSN: A Partition Density Clustering with Self-adaptive Neighborhoods

Published in Expert Systems With Applications, 2023

“PDCSN: A partition density clustering with self - adaptive neighborhoods” presents PDCSN. It uses self - adaptive methods to cluster, and outperforms rivals on multiple datasets.

Recommended citation:

PDCSN: A Partition Density Clustering with Self-adaptive Neighborhoods, S. Xing, Q.-M. Su*, Y.-J. Xiong*, C.-M. Xia,Expert Systems With Applications, 2023, 227 .1: 120195

Download Paper

40.2C2S: A Two-channel and Two-stream Transformer Based Framework for Offline Signature Verification

Published in Engineering Applications of Artificial Intelligence, 2023

The paper “2C2S: A two-channel and two-stream transformer based framework for offline signature verification” presents the 2C2S framework. It leverages a two - stream setup and special modules, outperforming rivals in signature verification.

Recommended citation:

2C2S: A Two-channel and Two-stream Transformer Based Framework for Offline Signature Verification, J.-X. Ren, Y.-J. Xiong*, H. Zhan and B. Huang, Engineering Applications of Artificial Intelligence, 2023, 118.1: 105639

Download Paper

39.Multiple Dependence Representation of Attention Graph Convolutional Network Relation Extraction Model

Published in IET Cyber-Physical Systems: Theory & Applications, 2023

This paper presents an MDR - GCN relation extraction model using multiple dependency tree representations and a GSF Extractor module, achieving good results on multiple datasets and analyzing relevant factors.

Recommended citation:

Multiple Dependence Representation of Attention Graph Convolutional Network Relation Extraction Model, L.-F. Zhao, Y.-J. Xiong*, Y.-B. Gao and W.-J. Yu, IET Cyber-Physical Systems: Theory & Applications, 2023, 9: 247-257

Download Paper

38.Deep Frame-Point Sequence Consistent Network for Handwriting Trajectory Recovery

Published in International Conference on Parallel and Distributed Systems, 2023

The paper “Deep Frame-Point Sequence Consistent Network for Handwriting Trajectory Recovery” presents a two - stream framework for handwriting trajectory recovery. It uses a module to synchronize training and shows good results in experiments.

Recommended citation:

Deep Frame-Point Sequence Consistent Network for Handwriting Trajectory Recovery, Y.-J. Xiong, Y.-F. Dai and D. Meng*, in Proceedings of the International Conference on Parallel and Distributed Systems, 2023: 2151-2158

Download Paper

36.两阶段问答范式的生物医学事件触发词检测

Published in 计算机工程与应用, 2024

This paper proposes a biomedical event trigger detection method based on a two - stage question - answering paradigm. It uses syntactic - distance - based attention and word - entity - event co - occurrence features to address existing problems in trigger detection. Experiments on the MLEE corpus show that the model outperforms baseline models, with an F1 - score of 81.39%, and the author also plans to explore further improvements in the future.

Recommended citation:

两阶段问答范式的生物医学事件触发词检测, 行帅, 熊玉洁*，苏前敏*, 黄继汉, 计算机工程与应用, 2024, 60.10: 21-131

Download Paper

35.Chain-of-LoRA: Enhancing the Instruction Fine-Tuning Performance of Low-Rank Adaptation on Diverse Instruction Set

Published in IEEE SIGNAL PROCESSING LETTERS, 2024

The paper proposes the Chain-of-LoRA framework, which trains a task - selection LoRA to classify instruction types and task - specific LoRAs for tasks. Experiments show it can achieve performance comparable to direct instruction fine - tuning, balancing performance and disk storage for resource - constrained users.

Recommended citation:

Chain-of-LoRA: Enhancing the Instruction Fine-Tuning Performance of Low-Rank Adaptation on Diverse Instruction Set, Qiu Xihe, Hao Teqi, Shi Shaojie, Tan Xiaoyu*,Xiong Yu-jie, IEEE Signal Processing Letters, 2024, 31: 875-879

Download Paper

34.Discrete Diffusion Models with Refined Language-image Pre-trained Representations for Remote Sensing Image Captioning

Published in Pattern Recognition Letters, 2024

The paper proposes DDM - RLIP, which applies a discrete diffusion model with refined pre - trained representations to remote sensing image captioning. Experiments on three datasets show it outperforms traditional autoregressive models.

Recommended citation:

Discrete Diffusion Models with Refined Language-Image Pre-trained Representations for Remote Sensing Image Captioning, Guannan Leng, Xiong Yu-jie*, Chunping Qiu*, Congzhou Guo, Pattern Recognition Letters, 2024, 186: 164-169

Download Paper

33.FaRE: A Feature-aware Radical Encoding Strategy for Zero-shot Chinese Character Recognition

Published in Asian Conference on Computer Vision, 2024

The paper proposes the FaRE strategy, which incorporates visual feature clues into radical encodings. Experiments on ICDAR2013 show it improves zero - shot Chinese character recognition performance compared to state - of - the - art methods.

Recommended citation:

FaRE: A Feature-aware Radical Encoding Strategy for Zero-shot Chinese Character Recognition, Zhan Hongjian, Li Yangfu, Xiong Yu-jie*, Lu Yue, Proceedings of the Asian Conference on Computer Vision, 2024, 390-401

Download Paper

32.基于Transformer特征通道融合的舌像分割

Published in 武汉大学学报（理学版）, 2024

In the field of tongue image segmentation, researchers have proposed a novel Transformer-based feature channel fusion method, which significantly improves segmentation accuracy and reliability.

Recommended citation:

基于Transformer特征通道融合的舌像分割, 薛玮珠*, 张博, 姚瑶, 熊玉洁, 夏春明, 武汉大学学报（理学版）, 2024, 70.6: 704-714

Download Paper

31.基于频繁模式挖掘算法的中医问诊策略研究

Published in 世界科学技术-中医药现代化, 2024

This paper uses the frequent pattern mining algorithm and cross - merging method to establish TCM single - and multi - system symptom questioning strategies, which can improve the efficiency of obtaining patient symptom information and promote the objective development of TCM consultation.

Recommended citation:

基于频繁模式挖掘算法的中医问诊策略研究, 李瑞珍, 夏春明*, 王忆勤, 许朝霞, 熊玉洁, 世界科学技术-中医药现代化, 2024, 26.6: 1608-1617

Download Paper

30.Free Lunch: Frame-level Contrastive Learning with Text Perceiver for Robust Scene Text Recognition in Lightweight Models

Published in ACM International Conference on Multimedia, 2024

This paper presents a frame - level contrastive learning framework with a Text Perceiver for lightweight scene text recognition models, improving performance, especially in low - quality scenarios, with effectiveness verified by experiments.

Recommended citation:

Free Lunch: Frame-level Contrastive Learning with Text Perceiver for Robust Scene Text Recognition in Lightweight Models, H.-J. Zhan, Y.-F. Li*, Y.-J. Xiong, Umapada Pal, Y. Lu, Proceedings of the 32nd ACM International Conference on Multimedia, 2024, 6202–6211

Download Paper

29.LRATNet: Local-Relationship-Aware Transformer Network for Table Structure Recognition

Published in International Conference on MultiMedia Modeling, 2024

“LRATNet: Local-Relationship-Aware Transformer Network for Table Structure Recognition” presents LRATNet. It combines modules for local and global info, a new loss function, and outperforms rivals on 3 datasets in table structure recognition.

Recommended citation:

LRATNet: Local-Relationship-Aware Transformer Network for Table Structure Recognition, G. Yang, D. Zhong, Y.-J. Xiong and H. Zhan*, in Proceedings of the International Conference on MultiMedia Modeling, Lecture Notes in Computer Science, 2024, 507--520

Download Paper

27.Multi-view Hypergraph Regularized Lp Norm Least Squares Twin Support Vector Machines for Semi-supervised Learning

Published in Pattern Recognition, 2024

The paper proposes MvHGLpLSTSVM for multi - view semi - supervised learning, combining hypergraph and Lp norm, and validates its effectiveness via experiments.

Recommended citation:

Multi-view Hypergraph Regularized Lp Norm Least Squares Twin Support Vector Machines for Semi-supervised Learning, J. Lu, X.-J. Xie* and Y.-J. Xiong, Pattern Recognition, 2024, 156: 110753

Download Paper

26.Enhanced Video Clustering Using Multiple Riemannian Manifold-valued Descriptors and Audio-visual Information

Published in Expert Systems With Applications, 2024

The paper proposes a method using multiple Riemannian manifold - valued descriptors and audio - visual information for video clustering, including single - and multi - modality approaches, and shows its superiority over existing methods through experiments.

Recommended citation:

Enhanced Video Clustering Using Multiple Riemannian Manifold-valued Descriptors and Audio-visual Information, W. Hu, H. Zhan*, Y. Tian, Y.-J. Xiong and Y. Lu, Expert Systems with Applications, 2024, 246: 123099

Download Paper

25.Transformer-based End-to-end Attack on Text CAPTCHAs with Triplet Deep Attention

Published in Computers & Security, 2024

The paper proposes a Transformer - based end - to - end method with triplet deep attention to attack text CAPTCHAs, achieving high accuracy on Roman and Chinese captcha datasets and exploring its performance under various conditions.

Recommended citation:

Transformer-based end-to-end attack on text CAPTCHAs with triplet deep attention, B. Zhang, Y.-J. Xiong*, C.-M. Xia and Y.-B. Gao, Computers & Security, 2024, 146: 104058

Download Paper

24.Towards Collaborative Intelligence: Propagating Intentions and Reasoning for Multi-Agent Coordination with Large Language Models

Published in arXiv, 2024

The paper proposes the REMALIS framework for multi - agent coordination with LLMs. It uses intention propagation, bidirectional feedback, and recursive reasoning, outperforming baselines.

Recommended citation:

Towards Collaborative Intelligence: Propagating Intentions and Reasoning for Multi-Agent Coordination with Large Language Models, X.-H. Qiu*, H.-Y. Wang, X.-Y. Tan, C. Qu, Y.-J. Xiong, Y. Chen, Y.-H. Xu, W. Chu, Y. Qi, arxiv preprint, arxiv:2407.12532 (2024)

Download Paper

23.PointABM: Integrating Bidirectional Mamba and Multi-Head Self-Attention for Point Cloud Analysis

Published in International Conference on Intelligent Technology and Embedded Systems, 2024

The paper proposes PointABM, which combines Bidirectional Mamba and Transformer, and it shows improved performance in point cloud analysis with a small increase in parameters.

Recommended citation:

PointABM: Integrating Bidirectional Mamba and Multi-Head Self-Attention for Point Cloud Analysis, J.-W. Chen, Y.-J. Xiong*, D.-H. Zhu, J.-C. Zhang, Z. Zhou, 2024 4th International Conference on Intelligent Technology and Embedded Systems. IEEE, 2024, 142-148

Download Paper

22.Kalman-SSM: Modeling Long-Term Time Series With Kalman Filter Structured State Spaces

Published in IEEE SIGNAL PROCESSING LETTERS, 2024

The paper “Kalman-SSM: Modeling Long-Term Time Series With Kalman Filter Structured State Spaces” presents the Kalman-SSM model. It combines the Kalman filter and SSM, outperforming SOTA models in long - term time series forecasting.

Recommended citation:

Kalman-SSM: Modeling Long-Term Time Series With Kalman Filter Structured State Spaces, Z. Zhou, X. Guo, Y.-J. Xiong* and C.-M. Xia, IEEE Signal Processing Letters, 2024, 31: 2470-2474

Download Paper

21.Harmonious Parameters and Performance: Lightweight Convolutional Stage and Local Feature Weighted fusion MLP for medical image segmentation

Published in Biomedical Signal Processing and Control, 2024

This paper proposes a lightweight medical image segmentation model named UConvNeXt based on depth - wise separable convolution and MLP. By using large - scale kernel depth - wise separable convolution and the local feature weighted fusion MLP (LFWF - MLP) module, experiments are carried out on multiple medical image datasets. The results show that while reducing parameters and computational complexity, the model can achieve comparable or even better segmentation performance than high - parameter models. Additionally, the limitations of the model and its future improvement directions are analyzed.

Recommended citation:

Harmonious Parameters and Performance: Lightweight Convolutional Stage and Local Feature Weighted fusion MLP for medical image segmentation, Y.-X. Chen, Y.-J. Xiong*, X.-H. Qiu and C.-M. Xia*, Biomedical Signal Processing and Control, 2024, 98: 106726

Download Paper

20.Adaptive Graph-based Feature Normalization for Facial Expression Recognition

Published in Engineering Applications of Artificial Intelligence, 2024

This paper presents AGFN for FER to handle data uncertainties. It uses a Poisson graph generator and GCN, and outperforms other methods, especially with mislabeled data.

Recommended citation:

Adaptive Graph-based Feature Normalization for Facial Expression Recognition, Y.-J. Xiong*, Q. Wang, Y.-T. Du and Y. Lu, Engineering Applications of Artificial Intelligence, 2024, 129: 107623

Download Paper

19.Few-Shot Named Entity Recognition with the Integration of Spatial Features

Published in Wuhan University Journal of Natural Sciences, 2024

A two - stage framework for few - shot NER is proposed. It uses multiscale convolution and an improved prototypical network, and outperforms baselines in experiments.

Recommended citation:

Few-Shot Named Entity Recognition with the Integration of Spatial Features, Z.-W. Liu, B. Huang*, C.-M. Xia, Y.-J. Xiong, Z.-S. Zhang, Y.-Q. Zhang, Wuhan University Journal of Natural Sciences, 2024, 29.2: 125-133

Download Paper

18.AutoGRN: An Adaptive Multi-channel Graph Recurrent Joint Optimization Network with Copula-based Dependency Modeling for Spatio-temporal Fusion in Electrical Power Systems

Published in Information Fusion, 2024

The paper proposes AutoGRN, which integrates an adaptive multi - channel framework and copula - based modeling for spatio - temporal fusion in power systems, outperforming benchmarks in multivariate prediction tasks.

Recommended citation:

AutoGRN: An Adaptive Multi-channel Graph Recurrent Joint Optimization Network with Copula-based Dependency Modeling for Spatio-temporal Fusion in Electrical Power Systems, H.-Y. Wang, X.-H. Qiu*, Y.-J. Xiong, X.-Y. Tan, Information Fusion, 2025, 117: 102836

Download Paper

17.Triplet Trustworthiness Validation with Knowledge Graph Reasoning

Published in Engineering Applications of Artificial Intelligence, 2025

This paper presents TTMNM, a triple - strategy - based model, to validate knowledge graph triplets. Experiments show it outperforms baselines and is applicable in industrial datasets.

Recommended citation:

Triplet Trustworthiness Validation with Knowledge Graph Reasoning, G. Zhang, Y.-J. Xiong*, J.-P. Hu, C.-M. Xia, Engineering Applications of Artificial Intelligence, 2025, 141: 109813

Download Paper

16.Understanding Before Reasoning: Enhancing Chain-of-Thought with Iterative Summarization Pre-Prompting

Published in arXiv, 2025

This paper proposes the Iterative Summarization Pre-Prompting (ISP²) method, which enhances the complex reasoning capabilities of large language models by adaptively extracting candidate information, rating the reliability of information pairs, and performing iterative summarization. Experiments show that this method can significantly improve model performance. Additionally, the paper analyzes the summarization steps and error sources of ISP².

Recommended citation:

Understanding Before Reasoning: Enhancing Chain-of-Thought with Iterative Summarization Pre-Prompting, D.-H. Zhu, Y.-J. Xiong*, J.-C. Zhang, X.-J. Xie, C.-M. Xia, arxiv preprint, arxiv:2501.04341 (2025)

Download Paper

15.Parameter-Efficient Fine-Tuning of Large Language Models via Deconvolution in Subspace

Published in International Conference on Computational Linguistics, 2025

This paper proposes DCFT, a method for parameter - efficient fine - tuning of large language models using deconvolution in subspace. It overcomes rank - one decomposition limitations, shows good performance with fewer parameters, and optimizes computational efficiency.

Recommended citation:

Parameter-Efficient Fine-Tuning of Large Language Models via Deconvolution in Subspace, J.-C. Zhang, Y.-J. Xiong*, C.-M. Xia, D.-H. Zhu, X.-H. Qiu, Proceedings of the 31st International Conference on Computational Linguistics, 2025, 3924-3935

Download Paper

14.PAST: Pairwise Attention Swin Transformer for Offline Signature Verification

Published in International Journal on Document Analysis and Recognition, 2025

This paper addresses challenges in signature verification, proposing a Pairwise Attention mechanism to enable bidirectional info exchange between reference and query signatures without extra temporal assumptions. Combined with Swin Transformer, it forms PAST, resolving input fusion issues and performing well on datasets. It also finds training background info in CEDAR impacts results significantly.

Recommended citation:

PAST: Pairwise Attention Swin Transformer for Offline Signature Verification, Y.-J. Xiong*, J.-X. Ren, D.-H. Zhu, X.-J. Xie, X.-H. Qiu, International Journal on Document Analysis and Recognition, 2025, 1-13

Download Paper

13.Sugar-Coated Poison: Benign Generation Unlocks Jailbreaking

Published in Conference on Empirical Methods in Natural Language Processing, 2025

The paper proposes SCP, a jailbreak attack that exploits "Defense Threshold Decay"—where prolonged benign generation reduces an LLM’s attention to input—enabling stealthy shifts to harmful outputs; it also introduces POSD, a part-of-speech–based defense.

Recommended citation:

Sugar-Coated Poison: Benign Generation Unlocks Jailbreaking, Y.-H. Wu, Y.-J. Xiong*, H. Zhang, J.-C. Zhang, Z. Zhou, in Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025(accepted)

Download Paper

12.Mdm-Dta: Message Passing Neural Network with Molecular Descriptors and Mixture of Experts for Drug-Target Affinity Prediction

Published in SSRN, 2025

The paper proposes MDM-DTA, a novel drug–target affinity prediction model that integrates molecular graphs, molecular descriptors, and protein semantic embeddings via a Mixture of Experts framework to achieve state-of-the-art performance.

Recommended citation:

Mdm-Dta: Message Passing Neural Network with Molecular Descriptors and Mixture of Experts for Drug-Target Affinity Prediction, Y. Dai, X.-Y. Tan, H.-Y. Wang, G.-C. Ma, Y.-J. Xiong, X.-H. Qiu*, Available at SSRN, 5315145

Download Paper

11.Why 1 + 1 < 1 in Visual Token Pruning: Beyond Naïve Integration via Multi-Objective Balanced Covering

Published in Advances in Neural Information Processing Systems, 2025

MoB is a training-free visual token pruning method for MLLMs that uses geometric covering theory to optimally balance prompt alignment and visual preservation, achieving high acceleration with minimal performance loss.

Recommended citation:

Why 1+ 1< 1 in Visual Token Pruning: Beyond Naive Integration via Multi-Objective Balanced Covering, Y.-F. Li, H.-J. Zhan*, T.-Y. Chen, Y.-J. Xiong, Q. L, Y. L, Advances in Neural Information Processing Systems, 2025(accepted)

Download Paper

10.MSA²: Multi-task Framework with Structure-aware and Style-adaptive Character Representation

Published in Proceedings of the IEEE/CVF international conference on computer vision, 2025

MoB is a training-free visual token pruning method for MLLMs that uses geometric covering theory to optimally balance prompt alignment and visual preservation, achieving high acceleration with minimal performance loss.

Recommended citation:

MSA²: Multi-task Framework with Structure-aware and Style-adaptive Character Representation, Y.-F. Li, H.-J. Zhan*, Q. Liu, L. Sun, Y.-J. Xiong, Y. L, Proceedings of the IEEE/CVF international conference on computer vision, 2025(accepted)

Download Paper

9.LoRA² : Multi-Scale Low-Rank Approximations for Fine-Tuning Large Language Models

Published in Neurocomputing, 2025

The paper proposes LoRA², which trains LoRAs on orthogonal planes, improves the importance score algorithm, and shows better performance than baselines in fine - tuning large language models with fewer parameters.

Recommended citation:

LoRA² :Multi-Scale Low-Rank Approximations for Fine-Tuning Large Language Models, J.-C. Zhang, Y.-J. Xiong*, X.-H. Qiu, D.-H. Zhu, C.-M. Xia, Neurocomputing, 2025, 650: 130859

Download Paper

8.CRGT-SA: An Interlaced and Spatiotemporal Deep Learning Model for Network Intrusion Detection

Published in Frontiers of Information Technology & Electronic Engineering, 2025

This paper proposes CRGT-SA, a novel spatiotemporal deep learning model that integrates CNN, LSTM, gated TCN, and a self-attention mechanism for intrusion detection. Experiments on the UNSW-NB15 and NSL-KDD datasets demonstrate its superior accuracy, F1-score, and generalization capability compared to existing methods.

Recommended citation:

CRGT-SA: an interlaced and spatiotemporal deep learning model for network intrusion detection, J. Chen, W.-X. Liu, X.-H. Qiu*, W.-J. Lv, Y.-J. Xiong, Frontiers of Information Technology & Electronic Engineering, 2025, 26.7: 1115--1130.

Download Paper

7.Multi-view Unsupervised Feature Selection Based on Graph Discrepancy Learning

Published in Neurocomputing, 2025

To tackle key limitations of unsupervised multi-view feature selection, this paper proposes GDFS, which fuses global-local graphs, reduces structural discrepancies, and uses low-rank tensor constraint/consensus clustering for feature selection, outperforming SOTA methods on six benchmarks.

Recommended citation:

Multi-view unsupervised feature selection based on graph discrepancy learning, Y.-W. Xu, X.-J. Xie*, X.-L. Jiang, Y.-J. Xiong, Neurocomputing, 2025, 656: 131487.

Download Paper

6.Multi-view Semi-supervised Feature Selection with Multi-order Similarity and Tensor Learning

Published in Neurocomputing, 2025

To address existing multi-view semi-supervised feature selection methods’ reliance on original data graph structures and their neglect of multi-order domain knowledge and cross-view high-order relations, this study proposes a method fusing multi-order similarity and tensor low-rank learning, which is solved via iteration and validated superior on multiple benchmarks.

Recommended citation:

Multi-view semi-supervised feature selection with multi-order similarity and tensor learning, H.-Y. Chen, X.-J. Xie*, Y.-J. Xiong, Neurocomputing, 2025, 657: 131573.

Download Paper

5.An Innovative Contrastive Learning Approach to Improve Image Recognition Robustness and Interpretability via Simulated Environmental Perturbations

Published in Engineering Applications of Artificial Intelligence, 2025

To tackle noise challenges and limitations of traditional image processing methods (e.g., poor generalization, distribution shifts), this paper proposes ERIEP, a contrastive learning strategy that identifies invariant visual features and enhances noise resistance/interpretability, outperforming SOTA baselines on CIFAR-10/100 and ImageNet-1K under perturbations.

Recommended citation:

An innovative contrastive learning approach to improve image recognition robustness and interpretability via simulated environmental perturbations, L.-J. Cheng, X.-H. Qiu*, X.-Y. Tan, H.-Y. Wang, Y.-J. Xiong, Engineering Applications of Artificial Intelligence, 2025, 159: 111619.

Download Paper

4.Multi-view Unsupervised Feature Selection with Unified Measurement of Consistency and Diversity

Published in Pattern Recognition, 2025

To address graph-based MvUFS methods’ failure to jointly consider multi-view consistency/diversity and neglect of higher-order correlations, this work proposes a unified framework integrating consistency-diversity measurement, consensus graph fusion, and low-rank tensor, outperforming SOTA algorithms.

Recommended citation:

Multi-view Unsupervised Feature Selection with Unified Measurement of Consistency and Diversity, S.-K. Xu, X.-J. Xie*, G.-Q. Chao, Y.-J. Xiong, Pattern Recognition, 2025, 112728.

Download Paper

3.CRA-U: Lightweight U-Net with Component Ranking Attention for Skin Lesion Segmentation

Published in Pattern Analysis and Applications, 2026

We propose CRA-U, a lightweight skin lesion segmentation framework integrating Component Ranking Attention for efficient global feature fusion with linear complexity, achieving SOTA performance with ~1/41 parameters of TransUNet.

Recommended citation:

CRA-U: lightweight U-Net with component ranking attention for skin lesion segmentation, Z.-P. Ji, Y.-X. Chen, Y. -J. Xiong*, X.-J. Xie, C. -M. Xia, Pattern Analysis and Applications, 2026, 29.1:22

Download Paper

2.Wavelet-Based MLP for Efficient Sensor Signal Measurement and Forecasting

Published in IEEE Transactions on Instrumentation and Measurement, 2026

Real-world time series measurement faces high complexity, noise sensitivity, and poor interpretability. We propose WaveMLP, a lightweight wavelet-MLP framework with high accuracy, efficiency, and interpretability for edge deployment.

Recommended citation:

Wavelet-Based MLP for Efficient Sensor Signal Measurement and Forecasting, Z. Zhou, Y. -J. Xiong*, C. -M. Xia, IEEE Transactions on Instrumentation and Measurement, 2026, 75:1-11

Download Paper

1.Gradient-Direction-Aware Density Control for 3D Gaussian Splatting

Published in International Conference on Learning Representations, 2026

This paper proposes GDAGS, a gradient-direction-aware density control method that mitigates over-reconstruction and over-densification in 3D Gaussian Splatting for superior rendering quality with compact scene representations.

Recommended citation:

Gradient-Direction-Aware Density Control for 3D Gaussian Splatting, Z. Zhou, Y.-J. Xiong*, C.-M. Xia, in Proceedings of the International Conference on Learning Representations, 2026(accepted)

Download Paper

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.