CV
Curriculum Vitae
Xiuying Wang
BSc Engineering student researching egocentric intelligence, VLM systems, and federated continual learning.
Education
Queen Mary University of London
Bachelor of Science (Engineering)
Beijing University of Posts and Telecommunications
Bachelor of Science (Engineering)
- Average Mark: 92/100
- GPA: 3.81/4.0 (Ranking Top 10%)
Publications
Lightweight Federated Incremental Learning via Decoupled Replay.
Xiuying Wang, Yichen Li, Hang Su, Gaozhuo Liu, Shiwei Li, Chuang Zhao, Jiangming Shi, Imran Razzak
Ego-R1: Chain-of-Tool-Thought for Ultra-Long Egocentric Video Reasoning.
S. Tian, R. Wang, H. Guo, P. Wu, Y. Dong, Xiuying Wang, Jingkang Yang, Hao Zhang, Hongyuan Zhu, Ziwei Liu
Feature Distillation is the Better Choice for Model-Heterogeneous Federated Learning.
Yichen Li, Xiuying Wang, Wenchao Xu, Haozhao Wang, Yining Qi, Jiahua Dong, Ruixuan Li
Under Review
Rehearsal-Free Statistical Prototype Regularization for Federated Incremental Learning.
Xiuying Wang, Yichen Li, Jiahua Cheng, Xiwei Liu, Chuang Zhao, Bo Liu, Imran Razzak
Heterogeneity-aware Distillation for Federated Continual Learning.
Gaozhuo Liu, Yichen Li, Xiuying Wang, Yulong Li, Chuang Zhao, Yankai Jiang, Imran Razzak
Research Experience
Multimedia Lab, Nanyang Technological University (MMLab-NTU)
Working closely with Jingkang Yang (Ph.D.) in Ziwei Liu's team · Singapore
- EgoLife: supported the construction and annotation of the EgoLife dataset, and helped develop EgoRAG, a retrieval-augmented generation system for long-form egocentric video understanding.
- Ego-R1 [2]: participated in developing a reasoning-agent framework and benchmark for long egocentric video understanding, focusing on tool-augmented reasoning and evaluation design.
- Visual Generation: surveyed benchmarks for visual generation tasks, summarizing recent advances, representative methods, evaluation protocols, and emerging research directions.
Intelligent and Distributed Computing Lab, HUST & GenMI Research Lab, MBZUAI
Working closely with Yichen Li (Ph.D. Student) · Wuhan, China
- Heterogeneous Federated Learning [3]: developed a feature-distillation framework to address model heterogeneity in federated learning.
- Federated Continual Learning: led research on frameworks that leverage lightweight features for decoupled replay [1] and statistical prototypes for regularization [4], mitigating catastrophic forgetting in distributed settings.
Working Experience
Synvo.ai & MMLab-NTU
Research Engineer Intern in Chen Change Loy's team · Singapore
- AI Copilot for Retail Scenarios: led the end-to-end development pipeline, including model training, agent construction, deployment, encryption, and system optimization, to build a VLM-powered agent system for retail stores with detection, segmentation, and ReID capabilities.
- PoC for a Smart Glass Company: contributed to the construction of long-horizon memory for egocentric streaming videos captured by always-on smart glasses in everyday scenarios.
- Security for VLMs: conducted research on feature inversion attacks against VLMs in egocentric scenarios, analyzing privacy risks and investigating potential attack pipelines and defense mechanisms.
Action Intelligence
Research Engineer Intern in Jingkang Yang's team · Shanghai, China
- Scalable Egocentric Data Collection for Embodied AI: designing and implementing a multimodal data-collection pipeline that integrates smart glasses, IMU sensors, AprilTag-based calibration, hand-pose estimation algorithms, and customized hardware to collect scalable egocentric interaction data for embodied intelligence research.
Research Interests
Egocentric Intelligence
Egocentric video understanding, long-horizon memory, multimodal data collection, and embodied interaction datasets.
VLM Systems
Vision-language models, agentic multimodal systems, RAG, and real-world deployment for interactive AI applications.
Federated Learning & Security
Model-heterogeneous federated learning, privacy-preserving learning, and security analysis.
Continual Learning
Federated continual learning and catastrophic forgetting mitigation.
Skills
- Programming: Python, C/C++, Java, MATLAB
- Deep Learning & Multimodal AI: PyTorch, Transformers, LLMs/VLMs, RAG, agentic multimodal systems
- Computer Vision & Egocentric Systems: Detection, segmentation, ReID, hand-pose estimation, AprilTag calibration, smart glasses/IMU data collection
- Machine Learning: Federated learning and continual learning
- Tools: Model deployment, distributed training, vector databases, Linux, Git, Docker