I am an MSc student in Scientific and Data Intensive Computing at University College London (UCL). I received my BSc in Data Science and Big Data Technology from Xi'an Jiaotong-Liverpool University (XJTLU).

My research focuses on building robust and practical AI systems at the intersection of computer vision, multi-modal learning, and medical image analysis. I am particularly interested in representation learning under noisy or incomplete data, and the real-world robustness of medical foundation models.

Artificial Intelligence Computer Vision Multi-modal Learning Medical Image Analysis Embodied AI
Publications
  • 2026
    DPFT: A Framework for Decomposed Probabilistic Forecasting with Transformers
    Yifang Wang, Boliang Yuan, Mengjie Mao, Xinyun Zhang, Yiheng Zhong, Xinkun Wang, Senwei Liang
    ICIC 2026 (International Conference on Intelligent Computing)
    Accepted Code
  • 2026
    SAIF: A Stability-Aware Inference Framework for Medical Image Segmentation with Segment Anything Model
    Ke Wu, Shiqi Chen, Yiheng Zhong, Hengxian Lin, Yingxue Su, Yifang Wang, Junhao Jin, Guangyu Ren
    IEEE BIBM 2026 (International Conference on Bioinformatics and Biomedicine)
    Accepted arXiv Code
  • 2026
    MedFM-Robust: Benchmarking Robustness of Medical Foundation Models
    Xiangxiang Cui, Yifang Wang, Lijie Hu, Tianjin Huang, Lu Yin
    MICCAI 2026 (Medical Image Computing and Computer Assisted Intervention)
    Accepted Code
  • 2025
    Robust Multimodal Learning for Ophthalmic Disease Grading via Disentangled Representation
    Yifang Wang, Xinkun Wang, Senwei Liang, Feilong Tang, Chengzhi Liu, Ming Hu, Chao Hu, Junjun He, Zongyuan Ge, Imran Razzak
    MICCAI 2025
    arXiv
  • 2025
    Beyond Words: AuraLLM and SignMST-C for Sign Language Production and Bidirectional Accessibility
    Yulong Li, Yuxuan Zhang, Feilong Tang, Ming Hu, Zhixiang Lu, Haochen Xue, Jianghao Wu, Mian Zhou, Kang Dang, Chong Li, Yifang Wang, Imran Razzak, Jionglong Su
    KDD 2026 Datasets and Benchmarks Track Poster
    arXiv
  • 2025
    HBLSIR: Information Diffusion Modeling on Inherent Public Opinion Structure, Multidimensional Emotion and Dynamic Susceptibility
    Simin Lai, Senwei Liang, Yifang Wang, Qingxia Li, Jionglong Su
    ICIC 2025
    Springer
Education
University College London (UCL) Sep 2025 – Sep 2026
MSc Scientific and Data Intensive Computing  ·  London, UK
Xi'an Jiaotong-Liverpool University (XJTLU) Sep 2020 – Jul 2025
BSc Data Science and Big Data Technology  ·  Suzhou, China
🏅 University Academic Achievement Award (Top 10%), 2023/2024
Research Experience
DPFT: A Framework for Decomposed Probabilistic Forecasting with Transformers
First Author Dec 2025 – Present
  • Proposed DPFT, a decomposed probabilistic forecasting framework integrating seasonal decomposition with a heteroscedastic Gaussian head and MC Dropout for aleatoric/epistemic uncertainty separation.
  • Demonstrated consistent point-forecast accuracy and substantially improved calibration over deterministic baselines on ETTh1 and ETTh2 benchmarks.
SAIF: A Stability-Aware Inference Framework for Medical Image Segmentation with Segment Anything Model
Research Assistant, Co-author  ·  UCL, XJTLU May 2026 – Aug 2026
  • Co-authored a training-free, plug-and-play inference framework that improves the robustness of SAM-based medical image segmentation without retraining.
  • Developed stability-consistency scoring to model prompt and threshold uncertainty, filter unstable candidates, and perform stability-weighted fusion.
  • Evaluated SAIF on four medical segmentation datasets, achieving average gains of 4–8% in mDice and 5–9% in mIoU.
MedFM-Robust: Benchmarking Robustness of Medical Foundation Models
Research Assistant, Second Author Dec 2025 – Mar 2026
  • Designed a robustness benchmark with 40 perturbation types across 8 imaging modalities for medical foundation models.
  • Evaluated 5 vision-language models (LLaVA-Med, MedGemma, Gemini-2.5-flash, GPT-4o-mini) and 2 segmentation models (MedSAM, SAM-Med2D) under 5 fine-tuning strategies.
  • Revealed that LoRA exhibits nearly 2× the performance degradation of full fine-tuning; medical-specific corruptions dominate segmentation failure.
Robust Multimodal Learning for Ophthalmic Disease Grading via Disentangled Representation
Co-first Author, Research Intern  ·  MBZUAI Dec 2024 – Mar 2025
  • Designed EPRL and DiLR modules for enhanced feature selection and modality disentanglement; integrated self-distillation for robustness under incomplete modality conditions.
  • Achieved state-of-the-art accuracy, AUC, and F1 on ophthalmic disease datasets with incomplete modalities.
Beyond Words: AuraLLM and SignMST-C for Sign Language Production and Bidirectional Accessibility
Co-author  ·  XJTLU, MBZUAI, Monash University Sep 2024 – Dec 2024
  • Contributed to AuraLLM and SignMST-C; assisted in building CNText2Sign and CNSign datasets with gloss mappings and video annotations.
  • Implemented LoRA fine-tuning and RAG retrieval; achieved state-of-the-art BLEU-4 scores on sign language translation benchmarks.
HBLSIR: Information Diffusion Modeling on Inherent Public Opinion Structure, Multidimensional Emotion and Dynamic Susceptibility
Second Author  ·  XJTLU, Fisk University Mar 2025 – May 2025
  • Co-developed the HBLSIR framework for modeling information diffusion via hierarchical Bayesian SIR processes with 11-class emotion-aware propagation.
  • Evaluated on the VISTA dataset; outperformed traditional SIR models in both cumulative and daily propagation simulation.
Awards & Honors
🥇 XJTLU Graduation Poster Exhibition — University-level First Prize Jul 2025
🥈 National University Students Competition of Intelligent Vehicles — National Second Prize 2024
🏅 University Academic Achievement Award (Top 10%) 2023/2024
Skills
Programming:  Python, C++, PyTorch
Research:  Representation & Multi-modal Learning · Bayesian & Generative Modeling · Semi-supervised Learning
Robotics:  ROS, SLAM, LiDAR/IMU/RGB-D sensor fusion