时空大数据智能与计算实验室 Big Data AI & Spatial-Temporal Computing Lab 哈尔滨工业大学（深圳）计算机科学与技术学院 School of Computer Science and Technology, HIT Shenzhen

研究成果

Research Outcomes

代表工作

Featured Works

时空基础模型Spatiotemporal Foundation Models

Map as a Prompt: Learning Multi-Modal Spatial-Signal Foundation Models for Cross-scenario Wireless Localization

Map as a Prompt: Learning Multi-Modal Spatial-Signal Foundation Models for Cross-scenario Wireless Localization

Map as a Prompt: Learning Multi-Modal Spatial-Signal Foundation Models for Cross-scenario Wireless Localization

ICLR 2026ICLR 2026

提出多模态无线定位基础模型 SigMap，以周期自适应掩码学习稳健的信号表示，并将三维地图作为轻量提示，实现面向未知环境的跨场景定位。

SigMap combines cycle-adaptive masked signal modeling with lightweight 3D map prompts to support accurate cross-scenario wireless localization in unseen environments.
链接Link 代码Code 数据Data

多模态时空智能与具身智能Multimodal Spatiotemporal Intelligence and Embodied Intelligence

DSTA4D: Content-aware Adaptive Decoupling for 4D Point Cloud Video Understanding

DSTA4D: Content-aware Adaptive Decoupling for 4D Point Cloud Video Understanding

DSTA4D: Content-aware Adaptive Decoupling for 4D Point Cloud Video Understanding

Pattern Recognition 2026Pattern Recognition 2026

聚焦 4D 点云视频理解，提升智能体对动态三维环境的感知与建模能力。

The study focuses on 4D point cloud video understanding, with the goal of improving intelligent agents’ perception and modeling capabilities in dynamic 3D environments.
链接Link 代码Code 数据Data
Key Technologies for Multimodal Embodied Intelligence in Autonomous Traffic Command and Control

面向自主式交通指挥与管控的多模态具身智能关键技术研发

Key Technologies for Multimodal Embodied Intelligence in Autonomous Traffic Command and Control

深圳市科技创新局重点产业研发项目Shenzhen Science and Technology Innovation Bureau Key Industry R&D Project

融合多模态感知、具身智能与交通管控技术，研究面向复杂交通环境的自主理解、推理和决策方法。

This project integrates multimodal perception, embodied intelligence, and traffic control to enable autonomous understanding, reasoning, and decision-making in complex traffic environments.
链接Link 代码Code 数据Data

时间序列分析Time Series Analysis

RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting

RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting

RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting

KDD 2026KDD 2026

提出两阶段预测框架 RePatch，通过熵引导的动态分块、时间交互建模与向量量化，学习紧凑、可迁移且可复用的离散时间表示，并在 14 个真实数据集上展现出领先的预测与跨场景泛化能力。

RePatch combines entropy-guided dynamic patching, temporal interaction modeling, and vector quantization to learn compact, transferable, and reusable discrete representations for time series forecasting.
链接Link 代码Code 数据Data

大语言模型Large Language Models

Cryptography-based privacy-preserving large language models: a lifecycle survey of frameworks, methods, and future directions

Cryptography-based privacy-preserving large language models: a lifecycle survey of frameworks, methods, and future directions

Cryptography-based privacy-preserving large language models: a lifecycle survey of frameworks, methods, and future directions

Artificial Intelligence Review 2026Artificial Intelligence Review 2026

系统梳理密码学驱动的大语言模型隐私保护研究，围绕数据选择、微调和推理等生命周期阶段归纳现有框架与方法，并总结未来研究方向。

This survey organizes cryptography-based privacy-preserving LLM research across data selection, fine-tuning, and inference, and outlines promising directions for future work.
链接Link 代码Code 数据Data

研究课题

Research Topics

面向车路云一体化的高性能通感算关键技术与设备研发Key Technologies and Equipment for High-performance Sensing, Communication, and Computing in Vehicle-Road-Cloud Integration
广东省重点领域研发计划Guangdong Provincial Key-Area Research and Development Program 2026.01–2028.12 参与Participant
面向自主式交通指挥与管控的多模态具身智能关键技术研发Key Technologies for Multimodal Embodied Intelligence in Autonomous Traffic Command and Control
深圳市科技创新局重点产业研发项目Shenzhen Science and Technology Innovation Bureau Key Industry R&D Project 2026.01–2028.12 参与（高校负责人）Participant (University Lead)
基于多智能体工作流的生产指挥决策支持技术研究与应用Research and Application of Production Command Decision Support Based on Multi-agent Workflows
南方电网深圳供电局委托开发项目Commissioned Project by Shenzhen Power Supply Bureau 2026.01–2027.07 主持Principal Investigator
基于自相关子图生成的时空图条件推演技术研究Research on Spatiotemporal Graph Conditional Reasoning Based on Autocorrelated Subgraph Generation
国家自然科学基金面上项目National Natural Science Foundation of China General Program 2025.01–2028.12 主持Principal Investigator
异质性时空大数据预训练基础模型技术研究Research on Pre-trained Foundation Models for Heterogeneous Spatiotemporal Big Data
广东省基础与应用基础研究基金面上项目Guangdong Basic and Applied Basic Research Foundation General Program 2025.01–2027.12 主持Principal Investigator
时空大数据挖掘与智能技术研究Research on Spatiotemporal Big Data Mining and Intelligence
深圳市高校稳定支持项目Shenzhen University Stable Support Program 2023.10–2026.09 主持Principal Investigator