Lijuan Sun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lijuan Sun has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Yutong Gao
Key Laboratory of Ethnic Language Intelligent Analysis and Security Governance, Ministry of Education, Minzu University of China, Beijing, China and Key Laboratory of Trustworthy Distributed Computing and Service (BUPT), Ministry of Education, Beijing, China
7 shared publications Guoshun Nan
School of Cyberspace Security Beijing University of Posts and Telecommunications
4 shared publications- Jingchen Wu
Key Laboratory of Trustworthy Distributed Computing and Service, Ministry of Education
4 shared publications - Congyan Lang
Beijing Jiaotong University, Beijing, China
3 shared publications - Xu Wu
Key Laboratory of Trustworthy Distributed Computing and Service (BUPT), Ministry of Education
3 shared publications - Yunchao Wei
Beijing Jiaotong University, Beijing, China
3 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17914-17920
Handwriting Classification via EEG-fNIRS Delay Awareness and Channel Interaction
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17914-17920
Open AccessHandwriting Classification via EEG-fNIRS Delay Awareness and Channel Interaction
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 75-82
Relation-Aware Human Activity Graphs for Insider-Risk Attribution from Heterogeneous Digital Interaction Streams
Digital workplaces continuously record human interactions through heterogeneous digital interaction streams, including logon, device, file, web, and e-mail events. These streams provide complementary views of user activity, but an anomaly score alone does not explain how actions form an event or which initiating pattern is most plausible. We present a human-centric, relation-aware activity-graph framework for insider-risk event attribution. It converts multi-stream activities into an abnormal behavior relationship graph, represents temporal adjacency and empirical conditional dependence with typed edges, abstracts historical anomalous graphs into labeled risk-cause prototypes, and ranks the prototypes with a shared relational graph convolutional network. Contextual language embeddings encode heterogeneous behavior labels, while the retrieved prototype provides an analyst-readable activity chain and associated risk consequences. Here, root cause denotes an initiating behavior pattern supported by a historical event chain rather than an interventional causal effect. On the synthetic CMU-CERT r6.2 dataset, the method obtains query–prototype matching precision, recall, and F1-score of 0.944, 0.912, and 0.928, respectively, together with Hit@3 of 0.952 and Hit@5 of 0.965. The results indicate that relation-aware graph modeling can support human-centric anomaly-event interpretation from heterogeneous digital interaction streams.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 75-82
Open AccessRelation-Aware Human Activity Graphs for Insider-Risk Attribution from Heterogeneous Digital Interaction Streams
Digital workplaces continuously record human interactions through heterogeneous digital interaction streams, including logon, device, file, web, and e-mail events. These streams provide complementary views of user activity, but an anomaly score alone does not explain how actions form an event or which initiating pattern is most plausible. We present a human-centric, relation-aware activity-graph framework for insider-risk event attribution. It converts multi-stream activities into an abnormal behavior relationship graph, represents temporal adjacency and empirical conditional dependence with typed edges, abstracts historical anomalous graphs into labeled risk-cause prototypes, and ranks the prototypes with a shared relational graph convolutional network. Contextual language embeddings encode heterogeneous behavior labels, while the retrieved prototype provides an analyst-readable activity chain and associated risk consequences. Here, root cause denotes an initiating behavior pattern supported by a historical event chain rather than an interventional causal effect. On the synthetic CMU-CERT r6.2 dataset, the method obtains query–prototype matching precision, recall, and F1-score of 0.944, 0.912, and 0.928, respectively, together with Hit@3 of 0.952 and Hit@5 of 0.965. The results indicate that relation-aware graph modeling can support human-centric anomaly-event interpretation from heterogeneous digital interaction streams.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 1
SinColor: Uncertainty-Guided Single-Step Diffusion for Image Colorization
Image colorization is a fundamental yet challenging task in computer vision, aiming to recover plausible and spatially coherent colors from grayscale images. Recent advancements in diffusion models have enabled significant progress in this field, yet existing methods predominantly rely on multi-step diffusion processes. While effective for generating high-frequency details, these approaches are suboptimal for colorization, as color information is inherently low-frequency, spatially smooth, and globally consistent. This mismatch leads to two critical limitations: (1) color artifacts and inconsistency due to excessive noise in the color space, and (2) high computational cost that hinders practical application. In this work, we propose a novel single-step diffusion framework for efficient and high-quality image colorization. We introduce a color uncertainty estimation (CUE) module to identify reliable and uncertain regions in the image, allowing the model to prioritize local certainty while reasoning about confused regions. To focus the model on low-frequency color generation, we directly encode the grayscale image into a latent representation, remove structural components in the output, and reconstruct the final image via efficient decoding. Extensive experiments on ImageNet, COCO-Stuff, and Extended COCO-Stuff demonstrate that our approach achieves state-of-the-art performance while reducing inference time by 98% and trainable parameters by 97% compared to leading multi-step diffusion methods. Our contributions include a systematic analysis of diffusion-based colorization, a lightweight yet effective uncertainty-aware framework, and comprehensive validation of its efficiency and effectiveness.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35, 1
SinColor: Uncertainty-Guided Single-Step Diffusion for Image Colorization
Image colorization is a fundamental yet challenging task in computer vision, aiming to recover plausible and spatially coherent colors from grayscale images. Recent advancements in diffusion models have enabled significant progress in this field, yet existing methods predominantly rely on multi-step diffusion processes. While effective for generating high-frequency details, these approaches are suboptimal for colorization, as color information is inherently low-frequency, spatially smooth, and globally consistent. This mismatch leads to two critical limitations: (1) color artifacts and inconsistency due to excessive noise in the color space, and (2) high computational cost that hinders practical application. In this work, we propose a novel single-step diffusion framework for efficient and high-quality image colorization. We introduce a color uncertainty estimation (CUE) module to identify reliable and uncertain regions in the image, allowing the model to prioritize local certainty while reasoning about confused regions. To focus the model on low-frequency color generation, we directly encode the grayscale image into a latent representation, remove structural components in the output, and reconstruct the final image via efficient decoding. Extensive experiments on ImageNet, COCO-Stuff, and Extended COCO-Stuff demonstrate that our approach achieves state-of-the-art performance while reducing inference time by 98% and trainable parameters by 97% compared to leading multi-step diffusion methods. Our contributions include a systematic analysis of diffusion-based colorization, a lightweight yet effective uncertainty-aware framework, and comprehensive validation of its efficiency and effectiveness.
- Open Access
Spotlighter: Revisiting Prompt Tuning from a Representative Mining View
CLIP's success has demonstrated that prompt tuning can achieve robust cross-modal semantic alignment for tasks ranging from open-domain recognition to fine-grained classification. However, redundant or weakly relevant feature components introduce noise and incur unnecessary computational costs. In this work, we propose Spotlighter, a lightweight token-selection framework that simultaneously enhances accuracy and efficiency in prompt tuning. Spotlighter evaluates each visual token's activation from both sample-wise and semantic-wise perspectives and retains only the top-scoring tokens for downstream prediction. A class-specific semantic memory bank of learned prototypes refines this selection, ensuring semantic representativeness and compensating for discarded features. To further prioritize informative signals, we introduce a two-level ranking mechanism that dynamically weights token--prototype interactions. Across 11 few-shot benchmarks, Spotlighter outperforms CLIP by up to 11.19\% in harmonic mean accuracy and achieves up to 0.8K additional FPS, with only 21 extra parameters. These results establish Spotlighter as an effective and scalable baseline for prompt tuning. Code for our method will be available at https://github.com/greatest-gourmet/Spotlighter.
Open AccessSpotlighter: Revisiting Prompt Tuning from a Representative Mining View
CLIP's success has demonstrated that prompt tuning can achieve robust cross-modal semantic alignment for tasks ranging from open-domain recognition to fine-grained classification. However, redundant or weakly relevant feature components introduce noise and incur unnecessary computational costs. In this work, we propose Spotlighter, a lightweight token-selection framework that simultaneously enhances accuracy and efficiency in prompt tuning. Spotlighter evaluates each visual token's activation from both sample-wise and semantic-wise perspectives and retains only the top-scoring tokens for downstream prediction. A class-specific semantic memory bank of learned prototypes refines this selection, ensuring semantic representativeness and compensating for discarded features. To further prioritize informative signals, we introduce a two-level ranking mechanism that dynamically weights token--prototype interactions. Across 11 few-shot benchmarks, Spotlighter outperforms CLIP by up to 11.19\% in harmonic mean accuracy and achieves up to 0.8K additional FPS, with only 21 extra parameters. These results establish Spotlighter as an effective and scalable baseline for prompt tuning. Code for our method will be available at https://github.com/greatest-gourmet/Spotlighter.
- 30 August 2025
- Published by Springer Nature
- p. 339-358
PDP-GAN-DD: Personalized Differential Privacy Generative Adversarial Network Algorithm for Data Desensitization
- 30 August 2025
- Published by Springer Nature
- p. 339-358
PDP-GAN-DD: Personalized Differential Privacy Generative Adversarial Network Algorithm for Data Desensitization
- 30 August 2025
- Published by Springer Nature
- p. 476-490
Rule-Based Learning for Explainable XGBoost Internal Threat Detection
- 30 August 2025
- Published by Springer Nature
- p. 476-490
Rule-Based Learning for Explainable XGBoost Internal Threat Detection
- 30 August 2025
- Published by Springer Nature
- p. 491-508
A Trusted Measurement Model for the Utilization of Medical Network Resources Based on User Behavior Portraits
- 30 August 2025
- Published by Springer Nature
- p. 491-508
A Trusted Measurement Model for the Utilization of Medical Network Resources Based on User Behavior Portraits
- 15 February 2025
- Published by Springer Nature
- p. 307-316
A Multi-holder Role and Strange Attractor-Based Data Possession Proof in Medical Clouds
- 15 February 2025
- Published by Springer Nature
- p. 307-316
A Multi-holder Role and Strange Attractor-Based Data Possession Proof in Medical Clouds
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 7435-7449
Spotlighter: Revisiting Prompt Tuning from a Representative Mining View
Yutong Gao, Maoyuan Shao, Xinyang Huang, Chuang Zhu, Yu Weng, Xuan Liu, Lijuan Sun, Guoshun Nan. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 7435-7449
Open AccessSpotlighter: Revisiting Prompt Tuning from a Representative Mining View
Yutong Gao, Maoyuan Shao, Xinyang Huang, Chuang Zhu, Yu Weng, Xuan Liu, Lijuan Sun, Guoshun Nan. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.
- 20 September 2024
- Published by Springer Nature
- p. 81-95
Generating a Personalised Sensor Data Generation System for the Fusion of Adversarial Networks and Behavioural Matter-of-Fact Mapping
- 20 September 2024
- Published by Springer Nature
- p. 81-95
Generating a Personalised Sensor Data Generation System for the Fusion of Adversarial Networks and Behavioural Matter-of-Fact Mapping