Alleviating Forgetting in Fine-Grained Urban Flow Inference via Incremental Neural Operator

In the realm of intelligent transportation, fine-grained urban flow inference (FUFI), which entails inferring fine-grained flow maps from their coarse-grained counterparts, has garnered considerable interest in the domain of sustainable urban traffic management. To tackle the FUFI, existing solutions primarily focus on exploring spatial dependencies, introducing external factors, reducing excessive memory costs, etc., while the challenge of effectively handling the diverse nature of urban traffic flow has not been fully addressed. These efforts rarely consider the catastrophic forgetting (CF) problem, which arises when previously learned knowledge is overwritten/forgotten as new knowledge is added. We propose an U rban N eural O perator solution with I ncremental learning ( UNOI ) inspired by recent advances in operator learning, which aims to learn grained-invariant solutions for FUFI in addition to addressing CF. Specifically, within the architecture of UNOI, we develop an urban neural operator (UNO) to capture spatial correlations more flexibly by treating the different-grained flows as continuous functions, learning mappings between approximation spaces more effectively. Furthermore, the phenomenon of CF behind time-related flows could hinder the capture of flow dynamics. Therefore, UNOI mitigates CF concerns as well as privacy issues by placing UNO blocks in two incremental settings, i.e., flow-related and task-related. In the end, our extensive experiments on large-scale real-world datasets demonstrate that our proposed solution consistently outperforms strong baselines.