姚鑫杰的照片

姚鑫杰Xinjie Yao

副教授Associate Professor
昆明理工大学
信息工程与自动化学院
Faculty of Information Engineering and Automation, Kunming University of Science and Technology

yaoxinjie@kust.edu.cn

个人简介About Me

姚鑫杰,昆明理工大学信息工程与自动化学院副教授,云南省人工智能重点实验室成员,所在团队由余正涛教授带领。2025 年获天津大学智能与计算学部博士学位,师从朱鹏飞教授、王煜教授和胡清华教授。Xinjie Yao (姚鑫杰) is an Associate Professor with the Faculty of Information Engineering and Automation at Kunming University of Science and Technology (KUST), and a member of the Yunnan Key Laboratory of Artificial Intelligence, led by Prof. Zhengtao Yu (余正涛). He received his Ph.D. in 2025 from the College of Intelligence and Computing, Tianjin University (TJU), under the supervision of Prof. Pengfei Zhu (朱鹏飞), Prof. Yu Wang (王煜), and Prof. Qinghua Hu (胡清华).

已发表学术论文 20 余篇,包括 ICML、NeurIPS、AAAI、IJCV、IEEE TMM、IEEE TCYB、PR 等重要会议与期刊。研究聚焦端到端自动化科研系统(AI4R)、人工智能驱动的科学研究(AI4S)与智能无人系统,重点探索人工智能驱动的腐蚀分析与预测、面向材料科学的多模态数据融合,以及适用于真实场景、尤其是低空环境的可扩展与可靠智能无人系统。His research has resulted in over 20 publications, including multiple papers in premier venues such as ICML, NeurIPS, AAAI, IJCV, IEEE TMM, IEEE TCYB, and PR. His work lies at the intersection of end-to-end automated research systems (AI4R), AI for science (AI4S), and intelligent unmanned systems. In particular, he focuses on AI-driven corrosion analysis and prediction, as well as multimodal data fusion for materials science, while advancing scalable and reliable intelligent unmanned systems for real-world deployment, especially in low-altitude environments.

教育经历Education

研究方向Researches

围绕“如何让智能系统在复杂、动态与协作环境中持续学习并解决真实科学问题”,团队重点探索以下四个方向,贯通基础方法研究与实际应用。

Our research asks how intelligent systems can learn continuously in complex, dynamic and collaborative environments and help solve real scientific problems. We connect foundational methods with practical applications across four directions.

图像、文本与传感信号通过跨模态对齐与融合,形成统一表征并支持可靠决策的示意图
异构信息 → 对齐融合 → 可靠认知
01 / 多模态学习

多模态学习

研究图像、文本、热红外与传感信号等异构信息的统一表征、跨模态对齐与可信融合,重点关注模态缺失、跨视图冲突和不完备数据条件下的鲁棒学习。

多视图学习 · 不完整多模态学习 · 可信融合

代表工作:IJCV 2024 ↗
模型依次学习新任务,通过知识记忆与迁移保持旧知识并扩展能力的示意图
保留旧知 → 学习新知 → 持续进化
02 / 持续学习

持续学习

面向数据、类别与环境不断变化的开放世界,研究新旧知识的协调更新与迁移,缓解灾难性遗忘,探索可见光—热红外等跨模态场景下的类增量学习。

类增量学习 · 抗遗忘 · 跨模态知识迁移

代表工作:PR 2026 ↗
多个具有不同任务和知识的智能体通过双向交流与协作共同提升能力的示意图
个体学习 ↔ 知识交互 ↔ 群体共进
03 / 多智能体学习

多智能体学习

探索智能体之间的知识共享、协作学习与社会化共同进化机制,使不同任务、模型与数据资源下的个体通过互助获得提升,推动从单体优化走向群体智能。

社会化学习 · 跨任务协作 · 共同进化

代表工作:ICML 2025 ↗
实验观测、人工智能建模与预测验证构成反馈闭环,用于材料腐蚀分析与性能预测的示意图
科学数据 → 智能建模 → 预测与验证
04 / 人工智能驱动的科学研究

人工智能驱动的科学研究

将机器学习与科学机理、实验观测相结合,聚焦材料腐蚀分析、退化评估与性能预测;融合视觉、电化学与环境信息,探索可解释、可验证的数据驱动科学研究方法。

材料智能 · 腐蚀预测 · 可解释科学建模

代表工作:PMS 2026 ↗
Diagram: images, text and sensor signals are aligned and fused for reliable decisions
Heterogeneous signals → aligned fusion → reliable perception
01 / Multimodal Learning

Multimodal Learning

We study unified representations, cross-modal alignment and reliable fusion across images, text, thermal imagery and sensor signals, focusing on robust learning with missing modalities, conflicting views and incomplete data.

Multi-view learning · Incomplete multimodal learning · Reliable fusion

Selected work: IJCV 2024 ↗
Diagram: a model learns successive tasks while retaining and transferring earlier knowledge
Retain prior knowledge → learn new tasks → continue adapting
02 / Continual Learning

Continual Learning

For open environments where data, classes and conditions evolve, we study how to update and transfer knowledge while mitigating catastrophic forgetting, including class-incremental learning across visible and thermal modalities.

Class-incremental learning · Forgetting mitigation · Cross-modal transfer

Selected work: PR 2026 ↗
Diagram: agents with distinct tasks and knowledge collaborate and improve together
Individual learning ↔ knowledge exchange ↔ collective progress
03 / Multi-Agent Learning

Multi-Agent Learning

We explore knowledge sharing, collaborative learning and socialized co-evolution among agents, enabling models with different tasks and data to improve through mutual assistance and collective intelligence.

Socialized learning · Cross-task collaboration · Co-evolution

Selected work: ICML 2025 ↗
Diagram: experiments, AI models and validation form a feedback loop for materials science
Scientific data → intelligent models → prediction and validation
04 / AI for Science

AI for Science

We combine machine learning with scientific mechanisms and experimental evidence to study materials corrosion, degradation assessment and performance prediction, integrating visual, electrochemical and environmental data.

Materials intelligence · Corrosion prediction · Explainable modeling

Selected work: PMS 2026 ↗

最新动态News

学术论文Publications

精选论文Selected Publications

  1. Socialized coevolution: advancing a better world through cross-task collaboration ICML
    Xinjie Yao, Yu Wang, Pengfei Zhu*, Wanyu Lin, Ruipu Zhao, Zhoupeng Guo, Weihao Li, Qinghua Hu.
    国际机器学习大会, 2025International Conference on Machine Learning, 2025
  2. Socialized learning: making each other better through multi-agent collaboration ICML
    Xinjie Yao#, Yu Wang#, Pengfei Zhu*, Wanyu Lin, Jialu Li, Weihao Li, Qinghua Hu.
    国际机器学习大会, 2024International Conference on Machine Learning, 2024
  3. Integrated heterogeneous graph attention network for incomplete multi-modal clustering IJCV
    Yu Wang, Xinjie Yao, Pengfei Zhu*, Weihao Li, Meng Cao, and Qinghua Hu.
    国际计算机视觉期刊, 2024International Journal of Computer Vision, 2024
  4. Incomplete cross-modality class-incremental learning in visible-thermal recognition PR
    Xinjie Yao, Yanxian Bi, Yu Wang*, Pengfei Zhu, Ruipu Zhao, Weihao Li, Wanyu Lin, Kai Li, Qinghua Hu.
    模式识别, 2026Pattern Recognition, 2026
  5. Multi-view deep subspace clustering networks TCYB
    Pengfei Zhu, Xinjie Yao, Yu Wang*, Binyuan Hui, Dawei Du, and Qinghua Hu.
    IEEE 控制论汇刊, 2024IEEE Transactions on Cybernetics, 2024
  6. Latent heterogeneous graph network for incomplete multi-view learning TMM
    Pengfei Zhu, Xinjie Yao, Yu Wang*, Meng Cao, Binyuan Hui, Shuai Zhao, and Qinghua Hu.
    IEEE 多媒体汇刊, 2022IEEE Transactions on Multimedia, 2022
  7. Visible-thermal cross-modality class-incremental learning ESWA
    Xinjie Yao, Yingxue Wang, Yu Wang*, Pengfei Zhu*, Ruipu Zhao, Weihao Li, Shenglei Pei, Wanyu Lin.
    专家系统及其应用, 2025Expert Systems with Applications, 2025
  8. Droc: open-set semi-supervised learning for robust perception in high-dimensional visual systems ICME
    Jin Li, Xinjie Yao*, Yu Wang, Pengfei Zhu.
    IEEE 国际多媒体与博览会, 2026International Conference on Multimedia and Expo, 2026
  9. Graphs help graphs: multi-agent graph socialized learning NeurIPS
    Jialu Li, Yu Wang, Pengfei Zhu, Wanyu Lin, Xinjie Yao, Qinghua Hu.
    神经信息处理系统大会, 2025Advances in Neural Information Processing Systems, 2025
  10. Zero-shot learning in industrial scenarios: new large-scale benchmark, challenges and baseline AAAI
    Zekai Zhang, Qinghui Chen, Maomao Xiong, Shijiao Ding, Zhanzhi Su, Xinjie Yao, Yiming Sun, Cong Bai, Jinglin Zhang.
    AAAI 人工智能大会, 2025AAAI Conference on Artificial Intelligence, 2025

按年份列出的论文 · GB/T 7714Publications by Year · GB/T 7714

收录 News 中全部 14 篇论文,按正式出版年份倒序排列;已分配未来卷期的论文按卷期年份列出,尚未正式出版的论文按录用年份列出并标注状态。采用 GB/T 7714—2015 顺序编码制:超过三位作者列前三位后加 et al.;[J] 为期刊论文,[C] 为会议论文。中文译名仅供理解,规范引用保留原始题名。All 14 papers mentioned in News are listed below in descending order of formal publication year. Papers assigned to future volumes are grouped by the volume year; other accepted papers are grouped by acceptance year and marked accordingly. References follow GB/T 7714—2015: the first three authors are followed by et al.; [J] denotes journal articles and [C] conference papers. Chinese title translations are provided for reading; formal citations retain the original titles.

2027

  1. 学习解决跨视图冲突,实现可靠的多视图分类

    LIAO Y, HOU S, WANG H, et al. Learning to resolve cross-view conflict for reliable multi-view classification[J]. Pattern Recognition, 2027, 183: 114917. DOI: 10.1016/j.patcog.2026.114917.

2026

  1. 结合计算、实验与人工智能辅助方法研究海洋环境中铝合金的多尺度腐蚀退化

    JI Y, LIU Z, YAO X, et al. Multi-scale corrosion degradation of aluminum alloys in marine environment studied by computational, experimental, and artificial intelligence-assisted approaches[J]. Progress in Materials Science, 2026, 162: 101740. DOI: 10.1016/j.pmatsci.2026.101740.

  2. 基于可解释视觉—阻抗多模态零样本学习的有机涂层退化分级

    WU Y, WU S, LIU W, et al. Degradation grading of organic coatings via explainable vision–impedance multimodal zero-shot learning[J]. Corrosion Science, 2026, 269: 113980. DOI: 10.1016/j.corsci.2026.113980.

  3. 可见光—热红外识别中的不完整跨模态类增量学习

    YAO X, BI Y, WANG Y, et al. Incomplete cross-modality class-incremental learning in visible-thermal recognition[J]. Pattern Recognition, 2026, 179: 113681. DOI: 10.1016/j.patcog.2026.113681.

  4. Droc:面向高维视觉系统鲁棒感知的开放集半监督学习

    LI J, YAO X, WANG Y, et al. Droc: open-set semi-supervised learning for robust perception in high-dimensional visual systems[C]//Proceedings of the IEEE International Conference on Multimedia and Expo. [S.l.]: IEEE, 2026.

2025

  1. 图助图:多智能体图社会化学习

    LI J, WANG Y, ZHU P, et al. Graphs help graphs: multi-agent graph socialized learning[C]//Advances in Neural Information Processing Systems 38. [S.l.]: Neural Information Processing Systems Foundation, 2025: 66079-66106. DOI: 10.52202/085713-1985.

  2. 基于语义分割与时间预测的海洋大气中划痕涂层/钢体系失效分析与腐蚀预测

    BAO Y, XIE G, PU P, et al. Failure analysis and corrosion prediction of scratched coating/steel system in marine atmosphere using semantic segmentation and temporal prediction[J]. Engineering Failure Analysis, 2025, 181: 109912. DOI: 10.1016/j.engfailanal.2025.109912.

  3. 可见光—热红外跨模态类增量学习

    YAO X, WANG Y, WANG Y, et al. Visible-thermal cross-modality class-incremental learning[J]. Expert Systems with Applications, 2025, 294: 128533. DOI: 10.1016/j.eswa.2025.128533.

  4. 社会化共同进化:通过跨任务协作迈向更好的世界

    YAO X, WANG Y, ZHU P, et al. Socialized coevolution: advancing a better world through cross-task collaboration[C]//Proceedings of the 42nd International Conference on Machine Learning: Vol. 267. [S.l.]: PMLR, 2025: 71780-71797.

  5. 工业场景中的零样本学习:全新大规模基准、挑战与基线

    ZHANG Z, CHEN Q, XIONG M, et al. Zero-shot learning in industrial scenarios: new large-scale benchmark, challenges and baseline[C]//Proceedings of the AAAI Conference on Artificial Intelligence: Vol. 39, No. 10. [S.l.]: AAAI Press, 2025: 10357-10366. DOI: 10.1609/aaai.v39i10.33124.

2024

  1. 社会化学习:通过多智能体协作实现彼此提升

    YAO X, WANG Y, ZHU P, et al. Socialized learning: making each other better through multi-agent collaboration[C]//Proceedings of the 41st International Conference on Machine Learning: Vol. 235. [S.l.]: PMLR, 2024: 56927-56945.

  2. 面向不完整多模态聚类的集成异构图注意力网络

    WANG Y, YAO X, ZHU P, et al. Integrated heterogeneous graph attention network for incomplete multi-modal clustering[J]. International Journal of Computer Vision, 2024, 132(9): 3847-3866. DOI: 10.1007/s11263-024-02066-y.

  3. 多视图深度子空间聚类网络

    ZHU P, YAO X, WANG Y, et al. Multiview deep subspace clustering networks[J]. IEEE Transactions on Cybernetics, 2024, 54(7): 4280-4293. DOI: 10.1109/TCYB.2024.3372309.

2023

  1. 面向不完整多视图学习的潜在异构图网络

    ZHU P, YAO X, WANG Y, et al. Latent heterogeneous graph network for incomplete multi-view learning[J]. IEEE Transactions on Multimedia, 2023, 25: 3033-3045. DOI: 10.1109/TMM.2022.3154592.

授权专利Patent

以下为中国授权发明专利,按 GB/T 7714 的专利文献格式列示;点击题名或“专利证书”可查看原始 PDF。The following granted Chinese invention patents are listed in GB/T 7714 patent-reference format. Select a title or certificate link to open the original PDF.
专利权人:天津大学。英文题名为译文,中文证书记载为准。Patent holder: Tianjin University. English titles are translations; the original Chinese certificates are authoritative.

  1. 姚鑫杰,王煜,朱鹏飞,等. 基于缺失跨模态数据类增量学习的无人机巡检方法及装置:CN119206552B[P]. 2026-05-08.YAO X, WANG Y, ZHU P, et al. UAV inspection method and apparatus based on class-incremental learning with incomplete cross-modal data: CN119206552B[P]. 2026-05-08.

  2. 姚鑫杰,王煜,朱鹏飞,等. 基于跨模态数据类增量学习的无人机巡检方法及系统:CN119131471B[P]. 2026-05-08.YAO X, WANG Y, ZHU P, et al. UAV inspection method and system based on class-incremental learning with cross-modal data: CN119131471B[P]. 2026-05-08.

  3. 朱鹏飞,姚鑫杰,曹萌,等. 一种基于集成异构图注意力网络的缺失多视图聚类方法:CN114565783B[P]. 2025-03-18.ZHU P, YAO X, CAO M, et al. Incomplete multi-view clustering method based on an integrated heterogeneous graph attention network: CN114565783B[P]. 2025-03-18.

科研项目Research Projects

主持或参与以下科研项目;经费为项目总经费。Selected projects led or participated in; funding denotes total project funding.

项目Project期限Period经费Funding状态Status角色Role
国家自然科学基金青年科学基金项目(C类)NSFC Young Scientists Fund (Category C)2027/01–2029/1230 万元 / CNY 300,000在研Ongoing主持Principal investigator
国家自然科学基金重点项目NSFC Key Program2025/01–2029/12227 万元 / CNY 2,270,000在研Ongoing参与Participant
天津市杰出青年项目Tianjin Outstanding Youth Program2023/10–2027/09100 万元 / CNY 1,000,000在研Ongoing参与Participant
科技创新2030—“新一代人工智能”重大项目Science and Technology Innovation 2030: New Generation Artificial Intelligence Major Project2023/03–2026/02500 万元 / CNY 5,000,000已结题Completed参与Participant
国家自然科学基金委员会优秀青年基金项目NSFC Excellent Young Scientists Fund2023/01–2025/12200 万元 / CNY 2,000,000已结题Completed参与Participant
科技部国家重点研发计划子课题Subproject of the National Key R&D Program of China2019/12–2022/11103 万元 / CNY 1,030,000已结题Completed参与Participant
国家自然科学基金委员会面上项目NSFC General Program2019/01–2022/1262 万元 / CNY 620,000已结题Completed参与Participant
国家自然科学基金委员会重点项目NSFC Key Program2018/01–2022/12285 万元 / CNY 2,850,000已结题Completed参与Participant
重庆长安汽车软件科技有限公司企业合作项目Industry collaboration with Chongqing Changan Automotive Software Technology Co., Ltd.2021/12–2022/1158 万元 / CNY 580,000已结题Completed参与Participant

团队学生Students

致有意报考的同学To Potential Students

欢迎有志于人工智能研究的同学报考昆明理工大学信息工程与自动化学院。我们期待以共同的问题意识为起点,在严谨求证与开放协作中,成长为能够独立思考、创造新知的研究者。

  1. 以兴趣为起点。联系前请阅读本主页的研究方向与代表论文,并查阅学校及学院主页,确认研究契合,说明您关注的问题与已有探索。
  2. 以指导支持成长。团队提供舒适、专注的科研环境与一对一指导,贯通问题凝练、方法创新、实验验证和学术表达,并乐于资助学生参与国内外学术交流。
  3. 以材料呈现积累。请将 PDF 简历发送至 yaoxinjie@kust.edu.cn,简述研究兴趣,可附成绩单、代表成果或代码。邮件标题:姓名-目前在读学校-硕士申请(或博士申请)。
  4. 以基础支撑探索。重视编程、数学与英语,尤其是论文阅读和学术写作;更珍视学术诚信、独立思考、主动沟通与持续精进的韧性。
  5. 以交流检验契合。完整材料通常在 1–2 周内完成初评。方向契合且有名额时,安排线上或线下面谈,欢迎坦诚交流任何关切;另给予 3 天准备指定论文分享,重点阐明挑战、科学问题、目标、贡献、方法与关键公式,并提出自己的判断。
  6. 博士申请重在科研潜力。另设 5–7 天基础科研能力测试,考察英文阅读、报告写作、文献检索、数学理解与归纳,以及发现问题、形成研究思路的能力。
  7. 尽早沟通,审慎选择。建议博士申请者提前联系,为研究准备与双向了解留足时间。招生以当年学校及学院政策、导师资格和实际名额为准。

我们不承诺轻松,承诺的是认真指导、坦诚反馈与共同成长。科研的价值不在于承受辛苦,而在于穿越不确定性,形成独立判断,创造值得信赖的新知。愿与您以好奇启程,以严谨求证,以韧性致远。

We welcome aspiring AI researchers to the Faculty of Information Engineering and Automation at Kunming University of Science and Technology. Through shared research questions, rigorous inquiry and open collaboration, we aim to help students become independent thinkers who contribute new knowledge.

  1. Start with research interest. Read our research areas and representative papers, and consult the university and faculty websites. Explain the questions that motivate you and the work you have done to explore them.
  2. Grow with support. We offer a comfortable, focused research environment and individual supervision from problem formulation and method development to experiments and academic communication. We are glad to support participation in academic exchanges in China and abroad.
  3. Show your preparation. Send a PDF résumé to yaoxinjie@kust.edu.cn with a brief account of your research interests. Transcripts, representative work and code are welcome. Suggested subject: Name-Current University-Master's Application (or PhD Application).
  4. Build strong foundations. We value programming, mathematics and English, particularly research reading and academic writing, alongside integrity, independent thought, proactive communication and the resilience to keep improving.
  5. Assess fit through discussion. Initial review normally takes 1–2 weeks after a complete application. Where interests align and places are available, we arrange an online or in-person interview and welcome candid questions. Applicants also receive three days to prepare an assigned paper presentation covering its challenges, research questions, goals, contributions, methods and key equations, together with their own assessment.
  6. PhD applicants: demonstrate research potential. An additional 5–7 day exercise assesses English reading, report writing, literature search, mathematical understanding and synthesis, and the ability to identify questions and develop research ideas.
  7. Reach out early; choose thoughtfully. PhD applicants are encouraged to make contact early to allow time for preparation and mutual understanding. Admission remains subject to annual university and faculty policies, supervisor eligibility and available places.

We do not promise an easy path. We promise careful supervision, honest feedback and a commitment to grow together. Research is valuable because it turns uncertainty into independent judgment and trustworthy knowledge. Let curiosity guide the questions, rigor shape the evidence, and perseverance sustain the journey.

软件著作权Software Copyrights

已授权软件著作权 10 项。Ten software copyrights have been registered.

  1. 致密油藏体积压裂下压力分析软件系统Pressure Analysis Software System for Volume Fracturing in Tight Oil Reservoirs2019SR0822107
  2. 致密油藏缝网条件下产能软件系统Productivity Software System for Fracture Networks in Tight Oil Reservoirs2019SR0822123
  3. 低渗油藏分段压裂水平井渗流场模拟软件系统Seepage-Field Simulation Software System for Multistage-Fractured Horizontal Wells in Low-Permeability Reservoirs2019SR0822138
  4. 低渗油藏分段压裂水平井参数优化软件系统Parameter Optimization Software System for Multistage-Fractured Horizontal Wells in Low-Permeability Reservoirs2019SR0822132
  5. 基于 .Net 的油气藏管理系统.NET-Based Oil and Gas Reservoir Management System2019SR0537808
  6. 基于 Android 的校园快递客户端Android-Based Campus Parcel Client2019SR0751809
  7. 基于 MATLAB 的浴缸温度监控系统MATLAB-Based Bathtub Temperature Monitoring System2019SR0751805
  8. 基于 Android 的新闻客户端Android-Based News Client2019SR0547216
  9. 基于 .Net 的远程监控管理系统.NET-Based Remote Monitoring and Management System2019SR0370888
  10. 基于科研文献数据库的信息挖掘系统Information Mining System Based on a Research Literature Database2016SR177183

荣誉奖励Honors

学术服务Services

联系方式Correspondence