Qian Xu

🧠 About Me

Hi there! I am currently a Ph.D. student in the School of Information at Renmin University of China. My research interests lie in high-performance data and AI systems, with a particular focus on vector databases, graph systems, GPU query processing, and efficient inference for diffusion language models. I am broadly interested in building practical systems that improve throughput, memory efficiency, and cost efficiency under real-world resource constraints.

🎓 Education

Renmin University of China, 2023.09 – Present
Ph.D. student in Computer Science
Renmin University of China, 2019.09 – 2023.06
B.Sc. in Computer Science

💼 Work Experience

Ant Group, Research Intern, 2025.02 – Present
Diffusion LLM AI Infra Department
Beijing HaiZhi XingTu Technology Co., Ltd, Research Intern, 2022.11 – 2025.01
Graph Computing Department
SenseTime, OCR Research Intern, 2021.11 – 2022.06

📄 Selected Publications

dInfer: An Efficient Inference Framework for Diffusion Language Models

Yuxin Ma, Lun Du, Lanning Wei, Kun Chen, Qian Xu, Kangyu Wang, et al.

Technical Report, 2025 Core Contributor

[Paper] [Project Page]

HARMONY: A Scalable Distributed Vector Database for High-Throughput Approximate Nearest Neighbor Search

Qian Xu, Feng Zhang, Chengxi Li, Lei Cao, Zheng Chen, Jidong Zhai, Xiaoyong Du

SIGMOD 2026 First Author

[Paper] [Project Page]

Tribase: A Vector Data Query Engine for Reliable and Lossless Pruning Compression using Triangle Inequalities

Qian Xu, et al.

SIGMOD 2025 First Author

[Paper] [Project Page]

Improving Graph Compression for Efficient Resource-Constrained Graph Analytics

Qian Xu, et al.

VLDB 2024 First Author

[Paper] [Project Page]

TensorSlim: Improving GPU Tensor Query Processing for Resource-Constrained Environments

Qian Xu, et al.

ICDE 2026 First Author

[Paper]

🛠️ Projects

dInfer: Efficient Inference Framework for Diffusion Language Models

[GitHub]

dInfer is a high-performance open-source inference framework for diffusion language models. It modularizes the dLLM inference pipeline into model execution, diffusion iteration management, decoding strategy, and KV-cache management. I contributed to system-level optimizations for efficient dLLM serving, especially KV-cache reuse for bidirectional diffusion language models.

Vicinity Refresh. Designed a sliding-window KV-cache refresh mechanism for bidirectional diffusion language models. Instead of naively recomputing or fully reusing all KV states, this method selectively refreshes tokens near the current decoding block while reusing stable remote cache states, improving inference efficiency while preserving generation quality.

🏆 Honors & Awards

CCF Young Talent Development Program, Ph.D. Student, 2025
National Graduate Scholarship, 2025
National Graduate Scholarship, 2024

📘 Patents

CN119759967A: ANN compression optimization for high-dimensional vectors, 2nd inventor
CN120086218A: Sub-neighbor network vector indexing, 2nd inventor
CN119169111A: Large-scale graph compression processing, 2nd inventor

🌍 Misc

Languages: English, CET-6

Last updated: June 2026