邓辉琦 Huiqi Deng

博士 助理教授
西安交通大学计算机科学与技术学院
智能网络与网络安全教育部重点实验室

denghq7@xjtu.edu.cn

Google Scholar

• 陕西省人才引进计划，2025

• 西安交通大学“青年优秀人才支持计划”，2024

• 国家资助博士后研究人员计划 B 档，2024（导师：管晓宏院士）

• 上海市超级博士后激励计划，2022

• 全国博士后创新创业大赛优胜奖，2021

研究兴趣：

（i）大模型与智能体可信安全

重点研究方向包括：LLM 可信推理、LLM 可解释性、LLM/Agent 安全评测与防护、LLM/Agent 数据生成。我们关注大模型和智能体在推理、规划、工具调用和数据生成过程中的可靠性与安全性，目标是构建更可解释、更可信、更可控的智能系统。

（ii）大模型交叉应用

重点研究方向包括：如 AI for Network Configuration、AI for Drug Discovery。我们关注如何用大模型、智能体和图机器学习解决网络系统、药物发现等应用中的真实问题，推动人工智能从通用模型走向高价值应用。

谷歌学术：https://scholar.google.com/citations?user=QEjqzXgAAAAJ&hl=en

工作经历：

• 2024 - 至今，西安交通大学 计算机科学与技术学院，助理教授

• 2022 - 2024，上海交通大学电子信息与电气工程学院，博士后

• 2019 - 2020，德州农工大学计算机系（TAMU），访问学者

• 2017 - 2019，香港浸会大学计算机系（HKBU），研究助理

长期招收硕士生、本科生：如果你对大模型、智能体、可信人工智能及其交叉应用感兴趣，具备较强的学习能力、自我驱动力和科研热情，且编程能力/数学基础突出，欢迎联系我们。Email: denghq7 at xjtu.edu.cn

（* 代表 equal contribution, # 代表通讯作者）
                      

         可解释、可信人工智能方向：

1. Huiqi Deng, Hongbin Pei, Quanshi Zhang, Mengnan Du. Attribution Explanations for Deep Neural Networks: A Theoretical Perspective. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. (IEEE TPAMI；CCF-A类期刊；IF = 18.6)

2. Huiqi Deng, Na Zou, Mengnan Du, Weifu Chen, Guocan Feng, Ziwei Yang, Zheyang Li, Quanshi Zhang. Unifying Fourteen Post-Hoc Attribution Methods With Taylor Interactions. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024 (IEEE TPAMI；CCF-A类期刊；IF = 18.6，详见知乎介绍：论统一14种输入重要性归因算法)

3. Huiqi Deng^*, Qihan Ren^*, Hao Zhang, Quanshi Zhang. Discovering and Explaining the Representation Bottleneck of DNNs. International Conference on Learning Representations, 2022 (ICLR Oral Paper 录用率1.6%；清华 A 类会议；详见知乎介绍：神经网络的知识表达瓶颈)

4. Dongrui Liu^*, Huiqi Deng^*, Xu Cheng, Qihan Ren, Kangrui Wang, Quanshi Zhang. Towards the difficulty for a deep neural network to learn concepts of different complexities. Conference and Workshop onNeural Information Processing Systems, 2023 (NeurIPS；CCF-A 类会议）

5. Qihan Ren^*, Huiqi Deng^*, Yunuo Chen, Siyu Lou, Quanshi Zhang. Bayesian Neural Networks Avoid Encoding Complex and Perturbation-Sensitive Concepts. International Conference on Machine Learning, 2023（ ICML; CCF-A类会议）

6. Huiqi Deng, Na Zou, Mengnan Du, Weifu Chen, Guocan Feng, Xia Hu. A Unified Taylor Framework for Revisiting Attribution Methods. AAAI Conference on Artificial Intelligence, 2021（AAAI; CCF-A类会议）

7. Huiqi Deng, Na Zou, Weifu Chen, Guocan Feng, Mengnan Du, Xia Hu. Mutual Information Preserving Back-propagation: Learn to Invert for Faithful Attribution. ACM SIGKDD Conference on Knowledge Discovery and Data Mining （KDD; CCF-A类会议）

8. Huiqi Deng, Weifu Chen, Qi Shen, Andy J. Ma, Pong C. Yuen, Guocan Feng. Invariant subspace learning for time series data based on dynamic time warping distance. Pattern Recognition, 2020(PR, CCF-B类期刊)

9. Huiqi Deng, Weifu Chen, Andy J. Ma, Qi Shen, Pong C. Yuen, Guocan Feng. Robust shapelets learning: Transform-invariant prototypes. Chinese Conference on Pattern Recognition and Computer Vision, 2018 (PRCV, CCF-C类会议)

10. Hongbin Pei, Huiyuan Li, Xufan Hou, Bo Yang, Huiqi Deng#. PolyGeom: Geometry-Aware Graph Transformer for Building Polygon Extraction in Remote Sensing Images. Remote Sensing. 
    

11. Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, Mengnan Du. Explainability for large language models: A survey. ACM Transactions on Intelligent Systems and Technology, 2024. (大模型可解释性综述，引用约 1500 次)

12. Huilin Zhou, Hao Zhang, Huiqi Deng, Dongrui Liu, Wen Shen, Shih-Han Chan, Quanshi Zhang. Explaining generalization power of a dnn using interactive concepts. AAAI Conference on Artificial Intelligence, 2024（AAAI; CCF-A类会议）

13. Jie Ren, Mingjie Li, Qirui Chen, Huiqi Deng, Quanshi Zhang. Defining and quantifying the emergence of sparse concepts in dnns. IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023. （CVPR; CCF-A类会议）
    
    
    AI 应用方向：

14. Hongbin Pei, Yu Li, Huiqi Deng, Jingxin Hai, Pinghui Wang, Jie Ma, Jing Tao, Yuheng Xiong, Xiaohong Guan. Multi-Track Message Passing: Tackling Oversmoothing and Oversquashing in Graph Learning via Preventing Heterophily Mixing. International Conference on Machine Learning, 2024（ ICML; CCF-A类会议；Spotlight Paper 录用率3.5%；算法应用在KDD CUP 2024 Task 2挑战赛，获得铜牌）  

15. Hongbin Pei, Taile Chen, Chen A, Huiqi Deng, Jing Tao, Pinghui Wang, Xiaohong Guan. HAGO-Net: Hierarchical Geometric Massage Passing for Molecular Representation Learning. AAAI Conference on Artificial Intelligence, 2024（AAAI; CCF-A类会议）

16. Qingxiong Tan, Andy J. Ma, Mang Ye, Baoyao Yang, Huiqi Deng, Vincent Wai-Sun Wong, et al. UA-CRNN: Uncertainty-aware convolutional recurrent neural network for mortality risk prediction.   ACM international conference on information and knowledge management, 2019 (CIKM, CCF-B类会议)