Yongjun Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yongjun Li has a total of 67 co-authors. Here are the top co-authors by number of shared publications:
Congjie Gao
School of Computer, Northwestern Polytechnical University, Xian, Shaanxi, China
5 shared publicationsJiaqi Yang
School of Computer, Northwestern Polytechnical University, Xian, Shaanxi, China
5 shared publicationsYinyin Zhang
School of Computer, Northwestern Polytechnical University, Xiàn, China
5 shared publications- Wenli Ji
School of Communication Engineering, Xian University of Science and Technology, Xian, Shaanxi, China
4 shared publications - Xiaowen Ma
School of Computer Science, Northwestern Polytechnical University, Shaanxi, China
4 shared publications You Peng
Northwestern Polytech Univ, Sch Comp, Xian 710072, Shaanxi, Peoples R China
4 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4524-4532
Geometry Matters: Angular Representations on the Hypersphere for Time Series Anomaly Detection
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 4524-4532
Open AccessGeometry Matters: Angular Representations on the Hypersphere for Time Series Anomaly Detection
- Open Access
- 19 October 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (9) , 920
FedVIB–AGP: defending against distributed backdoor attacks in federated learning via variational information bottleneck and activation-gap pruning
A distributed backdoor attack (DBA) decomposes a global trigger into local components injected by multiple compromised clients, allowing a strong assembled-trigger response to emerge in the global model after federated aggregation. Existing defenses primarily analyze submitted client updates or apply generic repair criteria, and ma22y therefore leave trigger-sensitive internal pathways in the aggregated model insufficiently addressed. We propose FedVIB–AGP, a trigger-aware post-aggregation repair framework for non-independent and non-identically distributed (non-IID) horizontal federated learning (FL) under DBA that preserves the standard Federated Averaging (FedAvg) pipeline. During benign local training, a variational information bottleneck (VIB) regularizer constrains stochastic latent representations and is intended to discourage the retention of redundant trigger-correlated information. At repair rounds, under a setting in which the server has access to non-test clean samples and an assembled-trigger transform, activation-gap targeted pruning (AGP) compares channel responses on paired clean and assembled-trigger inputs, masks channels exhibiting disproportionate trigger-induced amplification, and performs clean fine-tuning for utility recovery. Experiments on CIFAR-10, Fashion-MNIST, MNIST, and SVHN yield attack success rates (ASR) of 2.16, 2.56, 0.81, and 0.60%, respectively, while retaining clean accuracies of 73.27, 75.93, 92.45, and 90.38%. Relative to Certifiably Robust Federated Learning(CRFL), the lowest-ASR competing baseline in the reported comparisons, FedVIB–AGP reduces ASR by up to 10.29% and improves clean accuracy by up to 11.16%. These results demonstrate that coupling representation-level regularization with trigger-conditioned structural repair can effectively suppress DBA-induced backdoor behavior while preserving benign-task utility in federated learning.
- 19 October 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (9) , 920
Open AccessFedVIB–AGP: defending against distributed backdoor attacks in federated learning via variational information bottleneck and activation-gap pruning
A distributed backdoor attack (DBA) decomposes a global trigger into local components injected by multiple compromised clients, allowing a strong assembled-trigger response to emerge in the global model after federated aggregation. Existing defenses primarily analyze submitted client updates or apply generic repair criteria, and ma22y therefore leave trigger-sensitive internal pathways in the aggregated model insufficiently addressed. We propose FedVIB–AGP, a trigger-aware post-aggregation repair framework for non-independent and non-identically distributed (non-IID) horizontal federated learning (FL) under DBA that preserves the standard Federated Averaging (FedAvg) pipeline. During benign local training, a variational information bottleneck (VIB) regularizer constrains stochastic latent representations and is intended to discourage the retention of redundant trigger-correlated information. At repair rounds, under a setting in which the server has access to non-test clean samples and an assembled-trigger transform, activation-gap targeted pruning (AGP) compares channel responses on paired clean and assembled-trigger inputs, masks channels exhibiting disproportionate trigger-induced amplification, and performs clean fine-tuning for utility recovery. Experiments on CIFAR-10, Fashion-MNIST, MNIST, and SVHN yield attack success rates (ASR) of 2.16, 2.56, 0.81, and 0.60%, respectively, while retaining clean accuracies of 73.27, 75.93, 92.45, and 90.38%. Relative to Certifiably Robust Federated Learning(CRFL), the lowest-ASR competing baseline in the reported comparisons, FedVIB–AGP reduces ASR by up to 10.29% and improves clean accuracy by up to 11.16%. These results demonstrate that coupling representation-level regularization with trigger-conditioned structural repair can effectively suppress DBA-induced backdoor behavior while preserving benign-task utility in federated learning.
- 28 September 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 352
AMAD: AutoMasked Attention for unsupervised multivariate time series anomaly detection
- 28 September 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 352
AMAD: AutoMasked Attention for unsupervised multivariate time series anomaly detection
- 23 June 2026
- Published by Elsevier in Information Sciences
- Vol. 755
Knowledge annotation of multi-modal educational resources adaptable to new category discovery
- 23 June 2026
- Published by Elsevier in Information Sciences
- Vol. 755
Knowledge annotation of multi-modal educational resources adaptable to new category discovery
- 21 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP (4) , 1-14
A Multiview Trajectory Representation Learning Method for User Identification
With the development of location-based social networks, more and more users post their check-ins on social sites. Matching accounts on different social sites based on users’ spatiotemporal trajectories has become a hot issue in current research. Existing methods do not take into account the temporal characteristics, spatial characteristics, and sequential characteristics of the user’s trajectory at the same time, resulting in the inability to obtain accurate trajectory features. To solve this problem, we propose a new user identification approach based on the spatiotemporal characteristics of user trajectories, which consists of three parts: 1) convert trajectories into graph structures that preserve both spatial connectivity and access order, then use a graph attention network (GAT) to extract spatial representations—enabling effective modeling of users’ mobility structural patterns; 2) design a time-slice enhanced long short-term memory (LSTM) model to extract temporal periodicity and sequential features, which is robust to sparse trajectories with uneven time intervals; and 3) adopt a contrastive learning framework with a modified SimCLR to mine hard negative samples, solving the problem of positive–negative sample imbalance and enhancing the discriminability of trajectory representations. Extensive experiments on real datasets demonstrate that the proposed approach is superior to baseline methods in terms of performance and efficiency.
- 21 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP (4) , 1-14
A Multiview Trajectory Representation Learning Method for User Identification
With the development of location-based social networks, more and more users post their check-ins on social sites. Matching accounts on different social sites based on users’ spatiotemporal trajectories has become a hot issue in current research. Existing methods do not take into account the temporal characteristics, spatial characteristics, and sequential characteristics of the user’s trajectory at the same time, resulting in the inability to obtain accurate trajectory features. To solve this problem, we propose a new user identification approach based on the spatiotemporal characteristics of user trajectories, which consists of three parts: 1) convert trajectories into graph structures that preserve both spatial connectivity and access order, then use a graph attention network (GAT) to extract spatial representations—enabling effective modeling of users’ mobility structural patterns; 2) design a time-slice enhanced long short-term memory (LSTM) model to extract temporal periodicity and sequential features, which is robust to sparse trajectories with uneven time intervals; and 3) adopt a contrastive learning framework with a modified SimCLR to mine hard negative samples, solving the problem of positive–negative sample imbalance and enhancing the discriminability of trajectory representations. Extensive experiments on real datasets demonstrate that the proposed approach is superior to baseline methods in terms of performance and efficiency.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 13 (3) , 3336-3349
AlLNet: Alternative Learning is Better for Multimodal Fake News Detection
The significance of multimodal feature fusion in multimodal-based fake news detection is crucial. However, each modality’s contribution to the final detection outcome varies, and due to substantial differences in data distribution, a simple fusion approach is often inadequate. While existing works attempt to perform multimodal fusion using attention mechanisms, they still face the challenge of modal laziness—where a modality may be underutilized during training due to its low weight, resulting in insufficient learning. In this article, we propose the model alternative learning network (AlLNet). AlLNet addresses the issue of modal laziness by learning unimodal features separately during the training stage and employing an alternating independent training process. Additionally, it incorporates contrast learning to enhance the discriminative ability of image features. During the inference stage, AlLNet evaluates the uncertainty of different modalities using entropy and combines their predictions dynamically with appropriate weights to produce the final result. Moreover, AlLNet is designed to be lightweight and effectively handles modalities missing, making it well-suited for real-world applications. Our experiments conducted on three real-world datasets in English and Chinese demonstrate the advantages and effectiveness of our approach.
- 10 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 13 (3) , 3336-3349
AlLNet: Alternative Learning is Better for Multimodal Fake News Detection
The significance of multimodal feature fusion in multimodal-based fake news detection is crucial. However, each modality’s contribution to the final detection outcome varies, and due to substantial differences in data distribution, a simple fusion approach is often inadequate. While existing works attempt to perform multimodal fusion using attention mechanisms, they still face the challenge of modal laziness—where a modality may be underutilized during training due to its low weight, resulting in insufficient learning. In this article, we propose the model alternative learning network (AlLNet). AlLNet addresses the issue of modal laziness by learning unimodal features separately during the training stage and employing an alternating independent training process. Additionally, it incorporates contrast learning to enhance the discriminative ability of image features. During the inference stage, AlLNet evaluates the uncertainty of different modalities using entropy and combines their predictions dynamically with appropriate weights to produce the final result. Moreover, AlLNet is designed to be lightweight and effectively handles modalities missing, making it well-suited for real-world applications. Our experiments conducted on three real-world datasets in English and Chinese demonstrate the advantages and effectiveness of our approach.
- 10 February 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 313
DSAD: A dual-stream aligned detection framework for LLM-generated exam answers
- 10 February 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 313
DSAD: A dual-stream aligned detection framework for LLM-generated exam answers
- 11 November 2025
- Published by Elsevier in Neurocomputing
- Vol. 664
R2IS: Resilient and robust neighborhood rough feature selection using combination mutual information
- 11 November 2025
- Published by Elsevier in Neurocomputing
- Vol. 664
R2IS: Resilient and robust neighborhood rough feature selection using combination mutual information
- 9 June 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 324
TACSNet: Two-stage alignment cross-modal differential sensing network for concept prediction
- 9 June 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 324
TACSNet: Two-stage alignment cross-modal differential sensing network for concept prediction