SuDIS@ZJU

Data Drives, Efficiency Defines, Together We Compute.

SuDIS is the Sustainable Data Intelligence and Data Systems research group at Zhejiang University. We study how data quality, efficient models, and data systems can work together to make intelligent applications scalable, reliable, and deployable.

Core directions: Data-centric AI · Efficient AI · Data Systems for AI

Rapid · Reliable · Responsible · Resilient

Our Vision

SuDIS is a research group at Zhejiang University dedicated to resource-efficient, data-centric artificial intelligence and the data systems that make intelligent services reliable, scalable, and deployable. We investigate AI data preparation and quality governance, efficient inference and serving for large and multimodal models, spatiotemporal intelligence, model and system lightweighting, and data management for emerging LLM- and agent-driven workloads. Our work has appeared in SIGMOD, VLDB, KDD, WWW, NeurIPS, ICML, ICLR, ACL, and related venues, and is carried out with partners including Alibaba Cloud (ATH, PAI, and Database), Ant Group (healthcare, security, and computing platforms), Baidu’s foundation-model team, Tencent, ByteDance, Xiaohongshu, and OPPO. We train PhD and master’s students, undergraduate researchers, postdoctoral researchers, and interns through close mentorship and end-to-end research practice, supported by National Natural Science Foundation key, overseas-young-scholar, and young-scientist grants, National Key R&D Program young-scientist and major-subproject support, Zhejiang provincial major natural-science and Pioneer programs, and more than ten industry collaborations. We welcome students and interns who want to turn data intelligence and efficient computing into dependable systems with lasting scientific and societal value.

Latest News

SuDIS-ZJU Team
Jun 16, 2026 1 min read
SuDIS Participates in a Textbook Project on Data Engineering for Large Models

The team contributes to a textbook-development project on data engineering for general AI education.

SuDIS-ZJU Team
May 14, 2026 1 min read
MedMemoryBench Introduces a Multi-Turn Medical Dialogue Memory Benchmark

MedMemoryBench evaluates memory and reasoning in long, multi-turn medical conversations and releases associated resources.

SuDIS-ZJU Team
May 10, 2026 1 min read
Forcing-KV Releases Hybrid KV Cache Compression for Autoregressive Video Diffusion

Forcing-KV combines structural and similarity-based pruning to reduce KV-cache memory in autoregressive video diffusion models.

SuDIS-ZJU Team
Apr 11, 2026 1 min read
Huan Li Discusses Intelligent Workload Management for Cloud Data Warehouses

A PostgreSQL technology salon featured a discussion of intelligent workload management for cloud data warehouses.

See all posts

Latest Publications

Ankai Hao, Ke Chen, Huan Li, Lidan Shou (2026). LATTEArena: An Evaluation Framework for LLM-powered Tabular Feature Engineering (Extended Version). arXiv preprint.

Cite arXiv Code Project

Yihao Wang, Haoran Xu, Renjie Gu, Yixuan Ye, Xinyi Chen, Xinyu Mu, Yuan Gao, Chunxiao Guo, Peng Wei, Jinjie Gu, Huan Li, Ke Chen, Lidan Shou (2026). MedMemoryBench: Benchmarking Agent Memory in Personalized Healthcare. arXiv preprint.

Cite arXiv Code News

Yicheng Ji, Zhizhou Zhong, Jun Zhang, Qin Yang, XiTai Jin, Ying Qin, Wenhan Luo, Shuiyang Mao, Wei Liu, Huan Li (2026). Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models. arXiv preprint.

Cite arXiv Code Project

See all publications

Research, Projects and Training

SuDIS develops resource-efficient, data-centric AI and the systems that support it. Current work spans multimodal data quality, large-model inference and serving, data preparation, time-series intelligence, and next-generation data management. We connect rigorous research with open-source artifacts, real workloads, and responsible deployment, giving students and interns a complete path from problem formulation to reproducible systems and public communication.

Our public project portfolio highlights selected national, university, and industry collaborations. Project cards expose only the public project name, partner/source, and active period.

ProjectSourcePeriod
Continual post-training systems for reasoning models on domestic intelligent-computing platformsNational Key R&D Program2026–2029
Theory and methods for structured big-data foundation modelsNational research program2025–2028
Zhejiang University–Ant Joint Research Center for Big-Data Cognitive ComputingAnt Group2025–2028
Multimodal data governance for large-model trainingZhejiang Provincial Major Program2024–2026
Data preparation, optimization, and augmentation for domain large modelsZhejiang Provincial Innovation Program2024–2026
Data quality for IoT intelligenceNational Natural Science Foundation2023–2026
Multimodal data intelligence for new-energy vehiclesIndustry collaboration2025–2028
Continuous profiling for MoE inference optimizationCCF–Ant Research Fund2026–2027
Efficient inference for multimodal ultra-long sequencesCCF–Baidu Research Fund2025–2026
OLAP multi-tenant isolation and runtime resource optimizationCCF–Alibaba Cloud Research Fund2024–2025
AI-enabled chronic disease managementNational research program2026–2030
Personalized medical agents with memory layersIndustry collaboration2026–2027
Key technologies for edge-intelligent time-series imputationNational Natural Science Foundation2025–2027
Real-time big-data technology for modern industrial systemsMinistry of Education teaching case2025–2026

Join Us

We welcome doctoral students, postdoctoral researchers, research assistants, and undergraduate interns interested in:

  • LLM, multimodal LLM, and agentic inference optimization
  • Data preparation and quality governance for domain models
  • Data systems for multimodal, decentralized, and LLM-driven workloads
  • Intelligent computing for spatiotemporal data

The group supports research exchange and project practice with leading universities and technology companies.