Fast Few-Shot Graph Flow Prediction
Keywords: graph flow, flow prediction, graph neural network, few-shot learning, traffic prediction, neural network
TL;DR: We propose an efficient algorithm to perform few-shot graph flow prediction.
Abstract: Accurate prediction of traffic flow is crucial for optimizing transportation networks, mitigating congestion, and improving urban planning. However, existing approaches like graph neural networks (GNNs) and traffic simulations face challenges in predicting flow for unseen road networks without historical data. Without abundant training data, GNNs often generalize poorly to new graphs, while simulations can be computationally infeasible for large-scale networks. This paper tackles the problem of few-shot traffic flow prediction in unseen road networks. We propose a novel traffic simulation algorithm that efficiently predicts flow based on node and edge attributes. Through theoretical analysis, we demonstrate our approach closely approximates true flow with asymptotically optimal runtime complexity. Experiments on real-world road networks show our simulation algorithm outperforms GNNs for predicting traffic in unseen cities after training on only three cities. While motivated by traffic prediction in road networks, we expect our contributions to have broader applicability to general graph flow prediction problems across domains.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7883
Loading