Zaifeng Pan

Hi there! I’ve been a Ph.D. student at UC San Diego since fall 2024. I’m part of the PICASSO Lab, where I’m fortunate to be advised by Prof. Yufei Ding. My current research interests lie in machine learning systems, with a particular focus on making large language models (LLMs) more efficient from the perspectives of systems, compilers, and kernels, covering both inference and training. Recently, I have been focusing on systems optimization for LLM agents.

Before starting my Ph.D., I earned my master’s degree at Renmin University of China, where I was advised by Prof. Feng Zhang and worked closely with Dr. Zhen Zheng. My master’s research focused on compiler optimizations for GPU workloads, particularly industrial recommendation models. In retrospect, this work explored ideas now commonly associated with MegaKernels.

My Chinese name is 潘再峰. I also like being called 小再. If you’re interested in my research or just want to connect as a friend, feel free to email me at zapan@ucsd.edu.

Also, a shout-out to my roommate, who is an expert in computer architecture—check out his homepage and research!

Education

University of California, San Diego (UCSD), 2024 - Present
Ph.D. student in Computer Science & Engineering
Advisor: Prof. Yufei Ding
Renmin University of China (RUC), 2021 - 2024
M.S. in Computer Software and Theory
Advisor: Prof. Feng Zhang
Shanghai Jiao Tong University (SJTU), 2017 - 2021
B.E. in Mechanical Engineering

Work Experience

NVIDIA, DL Training Performance Intern, 2026
Mentor: Dr. Abhinav Goel
Amazon Web Services, Applied Scientist Intern, 2025
Mentors: Dr. Aninda Manocha and Dr. Zhuang Wang
Microsoft, Research Intern, 2023 - 2024
Mentor: Dr. Zhen Zheng
Alibaba PAI, Research Intern, 2021 - 2023
Mentor: Dr. Zhen Zheng

Selected Publications

arXiv

HiState: Hierarchical Prefix Cache Management for Hybrid LLM Serving in Agentic Workloads

Zaifeng Pan^*, Lucas Wang^*, Xinwei Qiang, Yanbo Zhou, Chang Chen, Zhengding Hu, Yue Guan, and Yufei Ding

arXiv preprint arXiv:2610.41508, 2026

Bib PDF

@article{pan2026histate,
  title = {HiState: Hierarchical Prefix Cache Management for Hybrid LLM Serving in Agentic Workloads},
  author = {Pan, Zaifeng and Wang, Lucas and Qiang, Xinwei and Zhou, Yanbo and Chen, Chang and Hu, Zhengding and Guan, Yue and Ding, Yufei},
  journal = {arXiv preprint arXiv:2610.41508},
  year = {2026}
}

arXiv

LeanSparse: Memory-Efficient Serving for Native Sparse-Attention LLMs

Zaifeng Pan^*, Chris Wu^*, Xinwei Qiang, Zhengding Hu, Zhongkai Yu, and Yufei Ding

arXiv preprint arXiv:2611.04671, 2026

Bib PDF

@article{pan2026leansparse,
  title = {LeanSparse: Memory-Efficient Serving for Native Sparse-Attention LLMs},
  author = {Pan, Zaifeng and Wu, Chris and Qiang, Xinwei and Hu, Zhengding and Yu, Zhongkai and Ding, Yufei},
  journal = {arXiv preprint arXiv:2611.04671},
  year = {2026}
}

arXiv

RR-Evict: Fine-Grained Prefix Cache Eviction beyond LRU for Agentic LLM Serving

Zaifeng Pan^*, Chris Wu^*, Zhengding Hu, Xinwei Qiang, Zhongkai Yu, and Yufei Ding

arXiv preprint arXiv:2609.32278, 2026

Bib PDF

@article{pan2026rrevict,
  title = {RR-Evict: Fine-Grained Prefix Cache Eviction beyond LRU for Agentic LLM Serving},
  author = {Pan, Zaifeng and Wu, Chris and Hu, Zhengding and Qiang, Xinwei and Yu, Zhongkai and Ding, Yufei},
  journal = {arXiv preprint arXiv:2609.32278},
  year = {2026}
}

arXiv

SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering

Zaifeng Pan, Qianxu Wang, Zhengding Hu, Chang Chen, Yue Guan, Yanbo Zhou, Steven Swanson, and Yufei Ding

arXiv preprint arXiv:2607.00151, 2026

Bib PDF

@article{pan2026smoothagent,
  title = {SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering},
  author = {Pan, Zaifeng and Wang, Qianxu and Hu, Zhengding and Chen, Chang and Guan, Yue and Zhou, Yanbo and Swanson, Steven and Ding, Yufei},
  journal = {arXiv preprint arXiv:2607.00151},
  year = {2026}
}

ICML’26

ScaleSim: Serving Large-Scale Multi-Agent Simulation with Invocation Distance-Based Memory Management

Zaifeng Pan^*, Yipeng Shen^*, Zhengding Hu, Zhuang Wang, Aninda Manocha, Zheng Wang, Zhongkai Yu, Yue Guan, and Yufei Ding

In Proceedings of the 43rd International Conference on Machine Learning, 2026

Bib PDF Code

@inproceedings{pan2026scalesim,
  title = {ScaleSim: Serving Large-Scale Multi-Agent Simulation with Invocation Distance-Based Memory Management},
  author = {Pan, Zaifeng and Shen, Yipeng and Hu, Zhengding and Wang, Zhuang and Manocha, Aninda and Wang, Zheng and Yu, Zhongkai and Guan, Yue and Ding, Yufei},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning},
  year = {2026}
}

NeurIPS’25

KVFlow: Efficient Prefix Caching for Accelerating LLM-Based Multi-Agent Workflows

Zaifeng Pan, Ajjkumar Patel, Yipeng Shen, Zhengding Hu, Yue Guan, Wan-Lu Li, Lianhui Qin, Yida Wang, and Yufei Ding

In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

Bib PDF Code

@inproceedings{pan2025kvflow,
  title = {KVFlow: Efficient Prefix Caching for Accelerating LLM-Based Multi-Agent Workflows},
  author = {Pan, Zaifeng and Patel, Ajjkumar and Shen, Yipeng and Hu, Zhengding and Guan, Yue and Li, Wan-Lu and Qin, Lianhui and Wang, Yida and Ding, Yufei},
  booktitle = {The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year = {2025}
}

MLSys’25

FastTree: Optimizing Attention Kernel and Runtime for Tree-Structured LLM Inference

Zaifeng Pan, Yitong Ding, Yue Guan, Zheng Wang, Zhongkai Yu, Xulong Tang, Yida Wang, and Yufei Ding

In Proceedings of Machine Learning and Systems, 2025

Bib PDF Code

@inproceedings{pan2025fasttree,
  title = {FastTree: Optimizing Attention Kernel and Runtime for Tree-Structured LLM Inference},
  author = {Pan, Zaifeng and Ding, Yitong and Guan, Yue and Wang, Zheng and Yu, Zhongkai and Tang, Xulong and Wang, Yida and Ding, Yufei},
  booktitle = {Proceedings of Machine Learning and Systems},
  year = {2025}
}

SC’24

RecFlex: Enabling Feature Heterogeneity-Aware Optimization for Deep Recommendation Models with Flexible Schedules

Zaifeng Pan, Zhen Zheng, Feng Zhang, Bing Xie, Ruofan Wu, Shaden Smith, Chuanjie Liu, Olatunji Ruwase, Xiaoyong Du, and Yufei Ding

In International Conference for High Performance Computing, Networking, Storage and Analysis, 2024

Bib PDF Code

@inproceedings{pan2024recflex,
  title = {RecFlex: Enabling Feature Heterogeneity-Aware Optimization for Deep Recommendation Models with Flexible Schedules},
  author = {Pan, Zaifeng and Zheng, Zhen and Zhang, Feng and Xie, Bing and Wu, Ruofan and Smith, Shaden and Liu, Chuanjie and Ruwase, Olatunji and Du, Xiaoyong and Ding, Yufei},
  booktitle = {International Conference for High Performance Computing, Networking, Storage and Analysis},
  pages = {1--15},
  year = {2024},
  organization = {IEEE}
}

ASPLOS’23

RECom: A Compiler Approach to Accelerating Recommendation Model Inference with Massive Embedding Columns

Zaifeng Pan, Zhen Zheng, Feng Zhang, Ruofan Wu, Hao Liang, Dalin Wang, Xiafei Qiu, Junjie Bai, Wei Lin, and Xiaoyong Du

In Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 4, 2023

🏆 Distinguished Artifact Award (presented at ASPLOS’24)

Bib PDF Code

@inproceedings{pan2023recom,
  title = {RECom: A Compiler Approach to Accelerating Recommendation Model Inference with Massive Embedding Columns},
  author = {Pan, Zaifeng and Zheng, Zhen and Zhang, Feng and Wu, Ruofan and Liang, Hao and Wang, Dalin and Qiu, Xiafei and Bai, Junjie and Lin, Wei and Du, Xiaoyong},
  booktitle = {Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 4},
  pages = {268--286},
  year = {2023}
}