Publications

2026

1. LLM Agent Memory: A Survey from a Unified Representation–Management Perspective

   系统梳理 LLM Agent 记忆的统一表示与管理框架，覆盖存储、检索、遗忘三大机制。 | A unified survey on LLM agent memory covering representation, management, and forgetting mechanisms.

   Zhenheng Tang, Xin He, Tiancheng Zhao, Fanjunduo Wei, Xiang Liu , et al.

   2026
2. ICLR CCF-A

   Ghost in the cloud: Your geo-distributed large language models training is easily manipulated

   首次揭示地理分布式 LLM 训练中的梯度投毒攻击漏洞，ICLR 2026。 | First to expose gradient poisoning vulnerabilities in geo-distributed LLM training; ICLR 2026.

   TANG Zichen, Zhenheng Tang, Gaoning Pan, Buhua Liu, Xin He , et al.

   In The Fourteenth International Conference on Learning Representations · 2026
3. ICLR-Workshop CCF-A

   Lang-PINN: From Language to Physics-Informed Neural Networks via a Multi-Agent Framework

   用多智能体 LLM 框架自动将自然语言问题转化为物理信息神经网络，ICLR 2026 Workshop Spotlight。 | Multi-agent LLM framework that converts natural language problems into physics-informed neural networks; ICLR 2026 Workshop Spotlight.

   Xin He, Liangliang You, Hongduan Tian, Bo Han, Ivor Tsang , et al.

   In ICLR 2026 Workshop on AI with Recursive Self-Improvement (Spotlight) · 2026
4. DAC

   ExpertFlow: Efficient Mixture-of-Experts Inference via Predictive Expert Caching and Token Scheduling

   联合优化 MoE 专家预测与 Token 调度的推理框架，显著降低 GPU 显存占用与 IO 开销，DAC 2026。 | Joint optimization of MoE expert prediction and token scheduling, cutting GPU memory and IO overhead; DAC 2026.

   Xin He, Shunkang Zhang, Kaijie Tang, Shaohuai Shi, Yuxin Wang , et al.

   2026

   PDF Bib

   @article{he2024expertflowoptimizedexpertactivation,
     pub_note = {联合优化 MoE 专家预测与 Token 调度的推理框架，显著降低 GPU 显存占用与 IO 开销，DAC 2026。 | Joint optimization of MoE expert prediction and token scheduling, cutting GPU memory and IO overhead; DAC 2026.},
     title = {ExpertFlow: Efficient Mixture-of-Experts Inference via Predictive Expert Caching and Token Scheduling},
     author = {He, Xin and Zhang, Shunkang and Tang, Kaijie and Shi, Shaohuai and Wang, Yuxin and Zeng, Zihao and Tang, Zhenheng and Chu, Xiaowen and Yin, Haiyan and Tsang, Ivor and Soon, Ong Yew},
     year = {2026},
     url = {https://arxiv.org/abs/2410.17954},
     google_scholar_id = {LkGwnXOMwfcC},
   }

5. Book

   动手学 AutoML：从 NAS 到大语言模型优化实战

   国内首本系统覆盖 NAS、HPO 到大模型推理优化的 AutoML 实战书籍，机械工业出版社，2026。 | First Chinese textbook systematically covering NAS, HPO, and LLM optimization in practice; China Machine Press, 2026.

   Xin He, Xiaowen Chu, Kaiyong Zhao, Qiang Wang, Zhenheng Tang , et al.

   2026

   贺鑫（新加坡科技研究局 A*STAR / 香港浸会大学 HKBU），褚晓文（香港科技大学广州校区 HKUST(GZ)），赵开勇（其域创新有限公司 Qiyu Innovation Ltd.），王强（哈尔滨工业大学（深圳）HIT Shenzhen），唐桢桁（香港科技大学 HKUST），董佩杰（香港科技大学广州校区 HKUST(GZ)）

   Bib

   @book{he2026automl,
     pub_note = {国内首本系统覆盖 NAS、HPO 到大模型推理优化的 AutoML 实战书籍，机械工业出版社，2026。 | First Chinese textbook systematically covering NAS, HPO, and LLM optimization in practice; China Machine Press, 2026.},
     title = {动手学 AutoML：从 NAS 到大语言模型优化实战},
     author = {He, Xin and Chu, Xiaowen and Zhao, Kaiyong and Wang, Qiang and Tang, Zhenheng and Dong, Peijie},
     publisher = {机械工业出版社 (China Machine Press)},
     year = {2026},
     month = jun,
     isbn = {9787111803645},
     note = {贺鑫（新加坡科技研究局 A*STAR / 香港浸会大学 HKBU），褚晓文（香港科技大学广州校区 HKUST(GZ)），赵开勇（其域创新有限公司 Qiyu Innovation Ltd.），王强（哈尔滨工业大学（深圳）HIT Shenzhen），唐桢桁（香港科技大学 HKUST），董佩杰（香港科技大学广州校区 HKUST(GZ)）},
   }

2025

1. arxiv

   GM-Skip: Metric-Guided Transformer Block Skipping for Efficient Vision-Language Models

   度量引导的 Transformer 块跳过策略，在视觉语言模型上实现高效推理且精度损失极小。 | Metric-guided block-skipping for efficient inference in vision-language models with minimal accuracy loss.

   Lianming Huang, Haibo Hu, Qiao Li, Xin He, Nan Guan , et al.

   arXiv preprint arXiv:2508.18227 · 2025

   Bib

   @article{huang2025gm,
     pub_note = {度量引导的 Transformer 块跳过策略，在视觉语言模型上实现高效推理且精度损失极小。 | Metric-guided block-skipping for efficient inference in vision-language models with minimal accuracy loss.},
     title = {GM-Skip: Metric-Guided Transformer Block Skipping for Efficient Vision-Language Models},
     author = {Huang, Lianming and Hu, Haibo and Li, Qiao and He, Xin and Guan, Nan and Xue, Chun Jason},
     journal = {arXiv preprint arXiv:2508.18227},
     google_scholar_id = {hqOjcs7Dif8C},
     year = {2025}
   }

2. arxiv

   AutoHete: An Automatic and Efficient Heterogeneous Training System for LLMs

   自动化异构 GPU 集群 LLM 训练系统，动态分配计算资源以最大化吞吐量。 | Automated heterogeneous GPU cluster training system for LLMs with dynamic resource allocation to maximize throughput.

   Zihao Zeng, Chubo Liu, Xin He, Juan Hu, Yong Jiang , et al.

   arXiv preprint arXiv:2503.01890 · 2025

   Bib

   @article{zeng2025autohete,
     pub_note = {自动化异构 GPU 集群 LLM 训练系统，动态分配计算资源以最大化吞吐量。 | Automated heterogeneous GPU cluster training system for LLMs with dynamic resource allocation to maximize throughput.},
     title = {AutoHete: An Automatic and Efficient Heterogeneous Training System for LLMs},
     author = {Zeng, Zihao and Liu, Chubo and He, Xin and Hu, Juan and Jiang, Yong and Huang, Fei and Li, Kenli and Lim, Wei Yang Bryan},
     journal = {arXiv preprint arXiv:2503.01890},
     google_scholar_id = {roLk4NBRz8UC},
     year = {2025}
   }

3. arxiv

   RouteMark: A Fingerprint for Intellectual Property Attribution in Routing-based Model Merging

   首个面向路由型模型融合的 IP 指纹方案，通过路由路径特征实现知识产权归属认证。 | First IP fingerprinting scheme for routing-based model merging, attributing ownership via routing path signatures.

   Xin He, Junxi Shen, Zhenheng Tang, Xiaowen Chu, Bo Li , et al.

   arXiv preprint arXiv:2508.01784 · 2025

   Bib

   @article{he2025routemark,
     pub_note = {首个面向路由型模型融合的 IP 指纹方案，通过路由路径特征实现知识产权归属认证。 | First IP fingerprinting scheme for routing-based model merging, attributing ownership via routing path signatures.},
     title = {RouteMark: A Fingerprint for Intellectual Property Attribution in Routing-based Model Merging},
     author = {He, Xin and Shen, Junxi and Tang, Zhenheng and Chu, Xiaowen and Li, Bo and Tsang, Ivor W and Ong, Yew-Soon},
     journal = {arXiv preprint arXiv:2508.01784},
     google_scholar_id = {Se3iqnhoufwC},
     year = {2025}
   }

4. KDD CCF-A

   BurstGPT: A Real-world Workload Dataset to Optimize LLM Serving Systems

   首个真实 LLM 服务突发请求数据集，揭示了现有调度策略在突发流量下的性能瓶颈，KDD 2025。 | First real-world LLM serving burst-request dataset, revealing scheduling bottlenecks under traffic spikes; KDD 2025.

   Yuxin Wang, Yuhan Chen, Zeyu Li, Xueze Kang, Zhenheng Tang , et al.

   2025

   PDF Code

2024

1. arxiv

   FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression

   基于自适应压缩的去中心化 LLM 训练系统，支持跨地理区域 GPU 协同训练。 | Decentralized LLM training across geo-distributed GPUs with adaptive gradient compression.

   Zhenheng Tang, Xueze Kang, Yiming Yin, Xinglin Pan, Yuxin Wang , et al.

   arXiv preprint arXiv:2410.12707 · 2024

   PDF
2. arxiv

   Fault-Tolerant Hybrid-Parallel Training at Scale with Reliable and Efficient In-memory Checkpointing

   大规模混合并行训练的内存检查点容错方案，在节点失效时实现快速恢复。 | In-memory checkpointing for fault-tolerant hybrid-parallel LLM training with fast recovery on node failures.

   Yuxin Wang, Xueze Kang, Shaohuai Shi, Xin He, Zhenheng Tang , et al.

   2024

   PDF

2023

1. MedAI

   MedPipe: End-to-End Joint Search of Data Augmentation Policy and Neural Architecture for 3D Medical Image Classification

   端到端联合搜索数据增强策略与神经网络结构，用于三维医疗图像分类，IEEE MedAI 2023。 | End-to-end joint NAS and data augmentation search for 3D medical image classification; IEEE MedAI 2023.

   Xin He, Xiaowen Chu

   IEEE MedAI · 2023

   PDF Code Bib

   @article{medpipe,
     pub_note = {端到端联合搜索数据增强策略与神经网络结构，用于三维医疗图像分类，IEEE MedAI 2023。 | End-to-end joint NAS and data augmentation search for 3D medical image classification; IEEE MedAI 2023.},
     title = {MedPipe: End-to-End Joint Search of Data Augmentation Policy and Neural Architecture for 3D Medical Image Classification},
     author = {He, Xin and Chu, Xiaowen},
     journal = {IEEE MedAI},
     year = {2023},
     publisher = {IEEE MedAI},
     google_scholar_id = {u-x6o8ySG0sC}
   }

2. AAAI CCF-A

   NAS-LID: Efficient Neural Architecture Search with Local Intrinsic Dimension

   利用局部本征维度（LID）量化子网几何特性，以更低内存开销实现准确的超网分区，AAAI 2023 CCF-A。 | Uses local intrinsic dimension to characterize subnet geometry for accurate supernet partition at lower memory cost; AAAI 2023.

   Xin He, Jiangchao Yao, Yuxin Wang, Zhenheng Tang, Cheung Ka Chun , et al.

   AAAI · 2023

   PDF Code Bib

   @article{NASLID,
     pub_note = {利用局部本征维度（LID）量化子网几何特性，以更低内存开销实现准确的超网分区，AAAI 2023 CCF-A。 | Uses local intrinsic dimension to characterize subnet geometry for accurate supernet partition at lower memory cost; AAAI 2023.},
     title = {NAS-LID: Efficient Neural Architecture Search with Local Intrinsic Dimension},
     author = {He, Xin and Yao, Jiangchao and Wang, Yuxin and Tang, Zhenheng and Ka Chun, Cheung and Simon, See and Han, Bo and Chu, Xiaowen},
     booktitle = {AAAI},
     year = {2023},
     journal = {AAAI},
     corr = {true},
     zhihu = {https://zhuanlan.zhihu.com/p/591027254},
     special = {CCF-A},
     gamma = {https://gamma.app/public/0t1h94j75bccc0q},
     dimensions = {true},
     google_scholar_id = {Tyk-4Ss8FVUC}
   }

3. IJCAI-LLM

   FusionAI: Decentralized Training and Deploying LLMs with Massive Consumer-Level GPUs

   利用消费级 GPU 去中心化部署超大规模 LLM 的框架，IJCAI 2023 LLM Workshop。 | Decentralized LLM training and deployment on massive consumer-grade GPUs; IJCAI 2023 LLM Workshop.

   Zhenheng Tang, Yuxin Wang, Xin He, Longteng Zhang, Xinglin Pan , et al.

   In IJCAI-LLM Workshop · 2023

   PDF Bib

   @inproceedings{tang2023fusionai,
     pub_note = {利用消费级 GPU 去中心化部署超大规模 LLM 的框架，IJCAI 2023 LLM Workshop。 | Decentralized LLM training and deployment on massive consumer-grade GPUs; IJCAI 2023 LLM Workshop.},
     title = {FusionAI: Decentralized Training and Deploying LLMs with Massive Consumer-Level GPUs},
     author = {Tang, Zhenheng and Wang, Yuxin and He, Xin and Zhang, Longteng and Pan, Xinglin and Wang, Qiang and Zeng, Rongfei and Zhao, Kaiyong and Shi, Shaohuai and He, Bingsheng and Chu, Xiaowen},
     year = {2023},
     corr = {true},
     journal = {IJCAI-LLM Workshop},
     booktitle = {IJCAI-LLM Workshop},
     google_scholar_id = {W7OEmFMy1HYC},
   }

2022

1. MICCAI CCF-B

   Evolutionary Multi-objective Architecture Search Framework: Application to COVID-19 3D CT Classification

   多目标进化 NAS 框架应用于新冠肺炎三维 CT 分类，兼顾精度与计算效率，MICCAI 2022 CCF-B。 | Multi-objective evolutionary NAS for COVID-19 3D CT classification balancing accuracy and efficiency; MICCAI 2022.

   Xin He, Guohao Ying, Jiyong Zhang, Xiaowen Chu

   In International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) · 2022

   PDF Code Bib

   @inproceedings{emars,
     pub_note = {多目标进化 NAS 框架应用于新冠肺炎三维 CT 分类，兼顾精度与计算效率，MICCAI 2022 CCF-B。 | Multi-objective evolutionary NAS for COVID-19 3D CT classification balancing accuracy and efficiency; MICCAI 2022.},
     title = {Evolutionary Multi-objective Architecture Search Framework: Application to COVID-19 3D CT Classification},
     author = {He, Xin and Ying, Guohao and Zhang, Jiyong and Chu, Xiaowen},
     booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI)},
     pages = {560--570},
     year = {2022},
     corr = {true},
     special = {CCF-B},
     zhihu = {https://zhuanlan.zhihu.com/p/542168478},
     dimensions = {true},
     google_scholar_id = {UeHWp8X0CEIC},
     special_old = {top premier international conferences for medical image analysis}
   }

2. ECCV Co-first author; CCF-B

   EAGAN: Efficient Two-stage Evolutionary Architecture Search for GANs

   两阶段进化架构搜索框架自动设计 GAN 网络结构，通讯作者，ECCV 2022 CCF-B。 | Two-stage evolutionary architecture search for GAN design; co-first author, ECCV 2022.

   Guohao Ying, Xin He, Bin Gao, Bo Han, Xiaowen Chu

   In ECCV · 2022

   PDF Code Bib

   @inproceedings{ying2021eagan,
     pub_note = {两阶段进化架构搜索框架自动设计 GAN 网络结构，通讯作者，ECCV 2022 CCF-B。 | Two-stage evolutionary architecture search for GAN design; co-first author, ECCV 2022.},
     title = {EAGAN: Efficient Two-stage Evolutionary Architecture Search for GANs},
     author = {Ying, Guohao and He, Xin and Gao, Bin and Han, Bo and Chu, Xiaowen},
     booktitle = {ECCV},
     corr = {true},
     zhihu = {https://zhuanlan.zhihu.com/p/542169585},
     special = {Co-first author; CCF-B},
     dimensions = {true},
     google_scholar_id = {qjMakFHDy7sC},
     year = {2022}
   }

3. ICML CCF-A

   Virtual Homogeneity Learning: Defending against Data Heterogeneity in Federated Learning

   虚拟同质化学习缓解联邦学习中的数据异质性问题，ICML 2022 CCF-A。 | Virtual homogeneity learning to mitigate data heterogeneity in federated learning; ICML 2022.

   Zhenheng Tang, Yonggang Zhang, Shaohuai Shi, Xin He, Bo Han , et al.

   In · 2022

   PDF Code Bib

   @inproceedings{tang2022virtual,
     pub_note = {虚拟同质化学习缓解联邦学习中的数据异质性问题，ICML 2022 CCF-A。 | Virtual homogeneity learning to mitigate data heterogeneity in federated learning; ICML 2022.},
     title = {Virtual Homogeneity Learning: Defending against Data Heterogeneity in Federated Learning},
     author = {Tang, Zhenheng and Zhang, Yonggang and Shi, Shaohuai and He, Xin and Han, Bo and Chu, Xiaowen},
     journal = {ICML},
     special = {CCF-A},
     corr = {true},
     google_scholar_id = {9yKSN-GCB0IC},
     year = {2022}
   }

2021

1. KBS CCF-C, 2700+ citations.

   AutoML: A survey of the state-of-the-art

   AutoML 领域最全综述，覆盖 NAS、HPO、数据增强等方向，KBS 2021，引用量超 2200。 | Comprehensive AutoML survey covering NAS, HPO, and data augmentation; KBS 2021, 2,700+ citations.

   Xin He, Kaiyong Zhao, Xiaowen Chu

   Knowledge-Based Systems · 2021

   PDF Website Bib

   @article{he2021automl,
     pub_note = {AutoML 领域最全综述，覆盖 NAS、HPO、数据增强等方向，KBS 2021，引用量超 2200。 | Comprehensive AutoML survey covering NAS, HPO, and data augmentation; KBS 2021, 2,700+ citations.},
     title = {AutoML: A survey of the state-of-the-art},
     author = {He, Xin and Zhao, Kaiyong and Chu, Xiaowen},
     journal = {Knowledge-Based Systems},
     volume = {212},
     pages = {106622},
     year = {2021},
     publisher = {Elsevier},
     special = {CCF-C, 2700+ citations.},
     zhihu = {https://zhuanlan.zhihu.com/p/158162306},
     corr = {true},
     dimensions = {true},
     google_scholar_id = {zYLM7Y9cAGgC}
   }

2. AAAI CCF-A

   Automated Model Design and Benchmarking of 3D Deep Learning Models for COVID-19 Detection with Chest CT Scans

   首次将神经架构搜索应用于胸部 CT 新冠肺炎检测，自动设计三维深度学习模型，AAAI 2021 CCF-A。 | First NAS application to COVID-19 chest CT detection, automating 3D deep learning model design; AAAI 2021.

   Xin He, Shihao Wang, Xiaowen Chu, Shaohuai Shi, J. Tang , et al.

   In AAAI · 2021

   PDF Bib

   @inproceedings{He2021AutomatedMD,
     pub_note = {首次将神经架构搜索应用于胸部 CT 新冠肺炎检测，自动设计三维深度学习模型，AAAI 2021 CCF-A。 | First NAS application to COVID-19 chest CT detection, automating 3D deep learning model design; AAAI 2021.},
     title = {Automated Model Design and Benchmarking of 3D Deep Learning Models for COVID-19 Detection with Chest CT Scans},
     author = {He, Xin and Wang, Shihao and Chu, Xiaowen and Shi, Shaohuai and Tang, J. and Liu, Xin and Yan, Chenggang Clarence and Zhang, Jiyong and Ding, Guiguang},
     booktitle = {AAAI},
     year = {2021},
     corr = {true},
     special = {CCF-A},
     dimensions = {true},
     google_scholar_id = {2osOgNQ5qMEC},
   }

2020

1. CCGRID

   Benchmarking the performance and energy efficiency of AI accelerators for AI training

   系统评测主流 AI 加速器在训练任务上的性能与能效，为硬件选型提供基准参考，CCGRID 2020。 | Systematic benchmark of AI accelerators on training performance and energy efficiency; CCGRID 2020.

   Yuxin Wang, Qiang Wang, Shaohuai Shi, Xin He, Zhenheng Tang , et al.

   In 2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID) Workshop · 2020

   Bib

   @inproceedings{wang2020benchmarking,
     pub_note = {系统评测主流 AI 加速器在训练任务上的性能与能效，为硬件选型提供基准参考，CCGRID 2020。 | Systematic benchmark of AI accelerators on training performance and energy efficiency; CCGRID 2020.},
     title = {Benchmarking the performance and energy efficiency of AI accelerators for AI training},
     author = {Wang, Yuxin and Wang, Qiang and Shi, Shaohuai and He, Xin and Tang, Zhenheng and Zhao, Kaiyong and Chu, Xiaowen},
     booktitle = {2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID) Workshop},
     pages = {744--751},
     year = {2020},
     corr = {true},
     organization = {IEEE},
     google_scholar_id = {d1gkVwhDpl0C}
   }

2. Benchmarking deep learning models and automated model design for COVID-19 detection with chest CT scans

   对主流深度学习模型在 COVID-19 检测任务上的系统性基准测试，MedRxiv 2020。 | Systematic benchmark of deep learning models for COVID-19 detection from chest CT scans; MedRxiv 2020.

   Xin He, Shihao Wang, Shaohuai Shi, Xiaowen Chu, Jiangping Tang , et al.

   MedRxiv · 2020

   PDF Bib

   @article{he2020benchmarking,
     pub_note = {对主流深度学习模型在 COVID-19 检测任务上的系统性基准测试，MedRxiv 2020。 | Systematic benchmark of deep learning models for COVID-19 detection from chest CT scans; MedRxiv 2020.},
     title = {Benchmarking deep learning models and automated model design for COVID-19 detection with chest CT scans},
     author = {He, Xin and Wang, Shihao and Shi, Shaohuai and Chu, Xiaowen and Tang, Jiangping and Liu, Xin and Yan, Chenggang and Zhang, Jiyong and Ding, Guiguang},
     journal = {MedRxiv},
     year = {2020},
     corr = {true},
     publisher = {Cold Spring Harbor Laboratory Press},
     zhihu = {https://zhuanlan.zhihu.com/p/147361922},
     google_scholar_id = {IjCSPb-OGe4C},
   }

2019

1. Big Data

   Computer-Aided Clinical Skin Disease Diagnosis Using CNN and Object Detection Models

   基于 CNN 与目标检测的皮肤病辅助诊断系统，IEEE Big Data 2019 Workshop。 | CNN and object detection based clinical skin disease diagnosis assistant; IEEE Big Data 2019.

   Xin He, Shihao Wang, Shaohuai Shi, Zhenheng Tang, Yuxin Wang , et al.

   In 2019 IEEE International Conference on Big Data (Big Data) Workshop · 2019

   PDF Bib

   @inproceedings{he2019computer,
     pub_note = {基于 CNN 与目标检测的皮肤病辅助诊断系统，IEEE Big Data 2019 Workshop。 | CNN and object detection based clinical skin disease diagnosis assistant; IEEE Big Data 2019.},
     title = {Computer-Aided Clinical Skin Disease Diagnosis Using CNN and Object Detection Models},
     author = {He, Xin and Wang, Shihao and Shi, Shaohuai and Tang, Zhenheng and Wang, Yuxin and Zhao, Zhihao and Dai, Jing and Ni, Ronghao and Zhang, Xiaofeng and Liu, Xiaoming and others},
     booktitle = {2019 IEEE International Conference on Big Data (Big Data) Workshop},
     pages = {4839--4844},
     year = {2019},
     corr = {true},
     organization = {IEEE},
     url = {https://ieeexplore.ieee.org/document/9006528},
     google_scholar_id = {eQOLeE2rZwMC},
   }