特聘序列

硕士生导师，深圳理工大学人工智能研究院特聘副教授，研究聚焦于多模态大语言模型，图像分析与计算机仿真，于人工智能顶级会议及交叉期刊发表论文20篇，成果发表于International Conference on Machine Learning (ICML)、IEEE Transactions on Medical Imaging (TMI)和Journal of Biomechanics等知名期刊会议。长期担任CVPR、Medical Image Analysis等国际顶级会议及期刊审稿人。拥有丰富的产学研经验，成功研发了行业领先的多模态基础模型，与上海AI Lab合作打造了全球首个医疗多模态基础模型群浦医，同时依托上海申康医院发展中心构建了国内最大的医疗大数据训练设施。研发成果已落地为在国际国内应用的AI产品SenseCare，并在国产化方面实现突破，与华为合作完成了AI系统的国产化迁移，攻克了CUDA至昇腾平台的关键技术。

多模态世界模型，图像处理和计算机仿真。专注于多模态世界模型的高效化、可信化及实用化研究，致力于解决其计算成本、知识融合与泛化能力三大核心挑战，推动前沿技术在高频场景中的高效落地与应用普惠。研究重点包括：

1)多模态大模型的高效计算与压缩:聚焦视觉令牌冗余与计算瓶颈，研究视觉令牌压缩、动态推理等高效计算范式，并通过蒸馏、量化与神经架构搜索实现模型轻量化，构建高性能、低延迟的多模态系统。

2)数据与知识双驱动的多模态学习:探索小样本环境下结构化知识与数据的协同机制，研究跨模态提示微调、参数高效适配等技术，实现知识引导的模型压缩与任务高效迁移。

3)多模态模型的可解释性与泛化性:针对模型决策黑盒与域外泛化难题，融合思维链与强化学习构建可信推理框架，研究不确定性建模与分布外适应机制，提升模型的透明度与跨场景稳健性。

学习经历

2013.4-2017.6，上海交通大学-日本千叶大学，博士

2010.9-2013.3，上海交通大学，硕士

2005.9-2010.7，浙江大学，学士

工作经历

2025.5-至今，深圳理工大学人工智能研究院，特聘副教授

2021.3-2025.4，香港中文大学博智感知交互研究中心，研究员

2019.8-2021.2，商汤科技/复星杏脉，高级研究员

2017.7-2019.7，上海交通大学，博士后

代表性论文

[1]Xiaoyu Yang, Lijian Xu*, Hongsheng Li, Shaoting Zhang. One Leaf Reveals the Season: Occlusion-Based Contrastive Learning with Semantic-Aware Views for Efficient Visual Representation. International Conference on Machine Learning2025.

[2]Xiaoyu Yang, Lijian Xu*, Simon Yu, Hongsheng Li, Shaoting Zhang. Segmentation and Vascular Vectorization for Coronary Artery by Geometry-based Cascaded Neural Network. IEEE Transactions on Medical Imaging. 2024.1:259-269.

[3]Lijian Xu, Fuyou Liang, Lixu Gu, Hao Liu. Flow instability detected in ruptured versus unruptured cerebral aneurysms at the internal carotid artery. Journal of Biomechanics. 2019. 72:187–199.

[4]Lijian Xu, Fuyou Liang, Hao Liu. Influence of aging-associated flow waveform variation on hemodynamics in aneurysms present at internal carotid artery: a computational model-based study. Computers in Biology and Medicine. 2018. 101:51–60.

[5]Lijian Xu, Lixu Gu, Hao Liu. Exploring potential association between flow instability and rupture in patients with matched-pairs of ruptured–unruptured intracranial aneurysms. Biomedical Engineering Online. 2016. 15: 461-477.

[6]Lijian Xu, Tianyang Yang, Lekang Yin, Y Vassilevski, Fuyou Liang. Numerical simulation of blood flow in aorta with dilation: a comparison between laminar and LES modeling methods. Computer Modeling in Engineering & Sciences. 2020. 124(2): 509-526.

[7]Lijian Xu, Fuyou Liang, Bing Zhao, Hao Liu. Morphological and hemodynamic factors associated with ruptured middle cerebral artery mirror aneurysms: a retrospective study. World Neurosurgery. 2020. 137: e138-e143.

[8]Lijian Xu, Michiko Sugawara, Gaku Tanaka, Hao Liu. Effect of elasticity on wall shear stress inside cerebral aneurysm at anterior cerebral artery. Technology and Health Care. 2016. 24(3): 349-357.

[9]Zhiqiang Zhang, Lijian Xu, Rong Liu, Fuyou Liang Importance of incorporating systemic cerebroarterial hemodynamics into com-putational modeling of blood flow in intracranial aneurysm. Journal of Hydrodynamics. 2020. 32:510-522.

[10]Xuanyu Li, Xiaosheng Liu, Lijian Xu, Fuyou Liang, Bing Zhao, Hao Liu. Tortuosity of the superficial femoral artery and its influence on blood flow patterns and risk of atherosclerosis. Biomechanics and Modeling in Mechanobiology. 2019, 18: 883-896.

[11]Xingdong Zhou, Lekang Yin, Lijian Xu, Fuyou Liang, Bing. Non-periodicity of blood flow and its influence on wall shear stress in the carotid artery bifurcation. Journal of Biomechanics. 2020, 101: 109617.

[12]Zhuowei Li, Qing Xia, Lijian Xu, Shaoting Zhang. A Deep Reinforced Tree-traversal Agent for Coronary Artery Centerline Extraction. MICCAI, 2021: 418-428

成果转化:

1.医学图像重建与诊断系统：开发血管分割及脑出血识别算法，首创基于深度学习的ASPECTS区域自动评分技术，脑卒中量化评估准确率达94.5%，通过NMPA二类证认证。

2.功能学参数计算系统：构建CT影像的血流动力学分析引擎，实现血管流速、FFR及脑灌注核心区AI定量，核心/半暗带体积测算误差<5%。

3.智能医学研究平台（上海申康）：研发支持多模态的医疗大数据训练框架，开发胸肺疾病多模态诊断系统，临床准确率93.2%。

4.国产化AI系统迁移（华为昇腾）：解决CUDA至昇腾Atlas300芯片的算子适配技术，识别准确率95.3%，获华为昇腾技术认证。

5.AI临床验证：在香港屯门医院建立肺结节AI双盲验证方案，敏感性达97.3%，形成香港医管局认证的申报标准。