Intelligent Spatiotemporal Computing

Sun, 08 Mar 2026 00:00:00 +0000

1. The Grand Vision

In the era of Cyber-Physical Systems (CPS) and Autonomous Mobility, understanding the physical world through spatiotemporal data (e.g., trajectories, sensor networks, traffic flows) is the cornerstone of urban intelligence. However, the real physical environment is highly dynamic, strictly resource-constrained, and fraught with uncertainty.

This research direction aims to develop next-generation Intelligent Spatiotemporal Computing paradigms. The ultimate goal is to bridge the fundamental gap between imperfect physical sensing and the urgent need for highly reliable, real-time, and resource-efficient decision-making in complex urban environments.

2. Key Research Questions

As a long-term research agenda, this project seeks to answer three overarching scientific questions in the field of Spatiotemporal AI:

RQ1: Robustness against Physical Uncertainty: How can AI models reliably perceive and reconstruct continuous spatiotemporal dynamics from highly fragmented, sparse, and noise-corrupted sensor streams?
RQ2: Efficiency and Scalability under Constraints: How do we design data-centric computing paradigms (e.g., active sampling, intelligent compression) that dynamically adapt to strict system budgets (bandwidth, edge computing power) without compromising downstream utility?
RQ3: Explainable and Causal Spatiotemporal Reasoning: Beyond black-box forecasting, how can we endow spatiotemporal models with the ability to disentangle complex physical mechanisms and provide causal attributions for safety-critical decisions?

3. Current Milestones & Publications

To address these grand challenges, my ongoing research has established several foundational frameworks, serving as the initial milestones for this agenda:

Tackling RQ1 (Data Recovery): We introduced a progressive chunked Transformer approach that effectively captures long-range dependencies and local topologies, significantly enhancing trajectory reconstruction from low-quality GPS data. (Published in Communications in Transportation Research)

Tackling RQ3 (Interpretability): We advanced the paradigm of spatiotemporal prediction by developing sequential decomposition and attribution mechanisms, enhancing both predictive accuracy and model transparency. (Published in Knowledge-Based Systems)

4. Future Impact

Looking forward, the methodologies developed in this direction will serve as the algorithmic infrastructure for broader applications, including Urban Digital Twins, Low-Altitude Economy (UAV routing), and Autonomous Fleet Management, driving the sustainable evolution of future smart cities.

Data-Centric AI |

Intelligent Spatiotemporal Computing

1. The Grand Vision

2. Key Research Questions

3. Current Milestones & Publications

4. Future Impact