Spatiotemporal prediction improves by masking ambiguous sensor data
Addressing Spatial Indistinguishability in Spatiotemporal Prediction via Optimal Transport-Guided Masking
Machine LearningArtificial Intelligence
Summary
Predicting future events based on data from sensors spread over space and time is hard when different sensors show similar past patterns but have very different futures. The authors present STOT, a method that helps computers better tell these sensors apart by focusing on uncertain or ambiguous parts of the data using a mathematical technique called optimal transport. This approach improves forecasting by using structures in the data that change over time and encourages the model to explore related contexts. Tests on multiple real-world datasets show that STOT matches or outperforms existing methods and helps explain its decisions.
What this means in practice
- •For sensor network operators: Improve forecasting accuracy in sensor networks by highlighting ambiguous data points during model training using structured masking guided by transport-based metrics.
- •For urban traffic management teams: Enhance traffic flow predictions by better capturing divergences in sensor data from similar historical patterns using the STOT masking strategy.
Authors
Guangyu Wang, Jiawei Tong
Abstract
Spatiotemporal prediction aims to learn discriminative representations from correlated temporal signals over spatial structures for accurate future inference. A central challenge is \emph{spatial indistinguishability}: different nodes may share similar historical patterns yet evolve toward divergent futures, severely degrading forecasting performance in real-world sensor networks. Existing embedding-based and graph neural network (GNN)-based approaches can partially detect such ambiguous nodes but rely on historical similarity, struggling to capture \emph{future behavioral divergence}. We propose \textbf{STOT} (\textbf{S}patio\textbf{T}emporal \textbf{O}ptimal \textbf{T}ransport), a self-supervised framework that resolves spatiotemporal ambiguity via structured masking guided by optimal transport. Our key idea treats indistinguishability as a \emph{disambiguation} problem: future states are inferred by exploiting concurrent spatial correlations and their time-varying similarity. We design a similarity-aware metric for dynamic inter-node relationships and an optimal transport-based masking strategy to emphasize ambiguous positions during pre-training. A batch consistency constraint preserves semantic coherence, while a random-walk masking mechanism promotes structured context exploration. Experiments on six real-world datasets show that STOT performs competitively with state-of-the-art baselines on the evaluated benchmarks and improved interpretability through transport-plan visualizations.