
吕新昱
新财经综合实验室
副教授、硕士生导师
个人主页:https://xinyulyu.github.io/
E-mail:xinyulyu68@gmail.com
西南财经大学柳林校区经世楼D205
※教师简介
吕新昱, 博士, 1277星际-XJ.1277.com|网站登陆副教授,硕导。2024年6月博士毕业于电子科技大学计算机科学与技术专业, 主要研究方向:大模型安全(越狱防御,幻觉治理),多模态学习。累积完成高水平学术论文20余篇。在中科院SCI-1区Top期刊 (如: TPAMI, TIP, IJCV), CCF-A类会议 (如: CVPR, NeurIPS) 上发表学术论文8篇 (第一/通讯作者论文7篇)。主持主研多项国家自然科学基金项目、科技部重大专项、四川省自然科学基金青年项目、四川省教育厅创新实验项目。获得四川省计算机学会优秀学生论文奖、全球智数教育创新大赛金奖。担任多个人工智能领域国际会议和期刊 (如:CVPR: Outstanding Reviewer 25, ICCV, NeurIPS, AAAI, TPAMI, TMM)审稿人。
尊重学生个性发展,以兴趣为驱动,在开放、平等的科研环境中开展学术探索。诚挚欢迎乐观勤奋、志存高远的同学联系交流!
※讲授课程
本科生:人工智能与现代科技、大数据可视化
硕士生:多模态大模型、多模态大模型原理与实践
博士生:多模态大模型前沿
※研究成果
代表性学术论文(*通讯作者):
[1] Lyu, Xinyu and Gao, Lianli* et al. Adaptive Fine-Grained Predicates Learning for Scene Graph Generation, IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2023, (中科院一区Top期刊,影响因子:23.6,财大外文A+).
[2] Lyu, Xinyu and Gao, Lianli et al. Multi-Concept Learning for Scene Graph Generation. IEEE Transactions on Image Processing (TIP), 2025. (中科院一区Top期刊,影响因子:11.6,财大外文A1).
[3] Gao, Lianli and Lyu, Xinyu* et al. Informative Scene Graph Generation via Debiasing. International Journal of Computer Vision (IJCV), 2025. (CCF A类期刊,影响因子:12.1,财大外文A1).
[4] Lyu, Xinyu and Chen, Beitao et al. Alleviating Hallucinations in Large Vision-Language Models through Hallucination-Induced Optimization, Proceedings of the 38th Conference on Neural Information Processing Systems (NeurIPS), 2024, (机器学习领域顶级会议,CCF A).
[5] Lyu, Xinyu and Gao, Lianli et al. Fine-Grained Predicates Learning for Scene Graph Generation. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, (计算视觉领域顶级会议,CCF A).
其他学术成果请见学术主页:Google Scholar
指导学生发表论文(†共同作者):
[1] Yuan, Shengming† and Lyu, Xinyu† et al. FlexAC: Towards Flexible Control of Associative Reasoning in Multimodal Large Language Models., Proceedings of the 39th Conference on Neural Information Processing Systems (NeurIPS), 2025 (机器学习领域顶级会议,CCF A,联合指导博士生袁胜明一作).
[2] Chen, Beitao and Lyu, Xinyu et al. SafePTR: Token-Level Jailbreak Defense in Multimodal LLMs via Prune-then-Restore Mechanism, Proceedings of the 39th Conference on Neural Information Processing Systems (NeurIPS), 2025, (机器学习领域顶级会议,CCF A,联合指导博士生陈邶涛一作).
[3] Lyu, Xinyu† and Chen, Beitao† et al. Alleviating Hallucinations in Large Vision-Language Models through Hallucination-Induced Optimization, Proceedings of the 38th Conference on Neural Information Processing Systems (NeurIPS), 2024, (机器学习领域顶级会议,CCF A,联合指导博士生陈邶涛共同一作).
[4] Zheng, Chaofan† and Lyu, Xinyu† et al. Prototype-based Embedding Network for Scene Graph Generation, IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, (计算视觉领域顶级会议,CCF A,联合指导硕士生郑超凡一作).
[5] Zheng, Chaofan and Gao, Lianli and Lyu, Xinyu et al. IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2023, (中科院一区Top期刊,财大外文A1,联合指导硕士生郑超凡一作).
[6] Wang, Shuang and Gao, Lianli and Lyu, Xinyu et al. Dynamic Scene Graph Generation via Temporal Prior Inference. Proceedings of the 30th ACM International Conference on Multimedia (ACM MM), 2022, (多媒体顶级会议,CCF A,联合指导硕士生王爽一作).
主要科研项目
[1] 四川省自然科学基金青年基金项目,2025ZNSFSC1463,开放世界下的视觉关系检测算法与应用研究,2025-01至2026-12, 在研,主持.
[2] 四川省教育厅省级普通本科高校创新性实验项目,多模态大模型驱动的普惠金融风控对话系统教学实践研究,2025-07至2027-07, 在研,主持.
[3] 科学技术部科技创新2030-"新一代人工智能"重大项目,2018AAA0102200,视听觉信息的协同认知、主动决策与平台验证,2023-01 至 2024-06,已结题,参与.
[4] 快手科技-高校科研项目,202109FKY00302,电商直播/视频的场景图生成,2020-01至2022-07,已结题,主研(排名第二).
[5] 国家自然科学基金,62020106008,协同视觉与语言处理的视觉自然认知关键技术研究,2021-01至2025-12,已结题,参与.