An Efficient Multi-Attention Graph Network (EMAGN) that linearises the spatial attention mechanism itself, inspired by the theory of fast high-dimensional Gaussian filtering, which surpasses Linformer and Performer in both accuracy and efficiency within the same backbone, owing to its traffic-network-aware adaptive clustering.
Abstract
Traffic forecasting is highly challenging due to complex and nonlinear spatial and temporal dependencies. Self-attention mechanisms have been widely adopted to model dynamic and long-range dependencies, achieving state-of-the-art performance, but suffer from limited scalability due to quadratic computational and memory complexity. To address this, we propose an Efficient Multi-Attention Graph Network (EMAGN) that linearises the spatial attention mechanism itself, inspired by the theory of fast high-dimensional Gaussian filtering. Two learned clustering matrices C_k and C_v adaptively group key and value vectors into M super-clusters, reducing complexity from O(N^2 d) to O(NMd) without sacrificing the flexibility of attention for dynamic dependency modelling. Experimental results on PEMS-BAY and METR-LA show that EMAGN achieves accuracy within 2.7-3.2% MAE of full-attention GMAN while reducing training time by 32%, inference time by 38%, and GPU memory by 58%. Critically, at K=16 attention heads, full-attention GMAN runs out of memory on a standard 11 GB GPU entirely while EMAGN continues to operate, demonstrating a categorical expansion of feasible model configurations. EMAGN also surpasses Linformer and Performer in both accuracy and efficiency within the same backbone, owing to its traffic-network-aware adaptive clustering.
Achieving accurate and scalable traffic prediction is a cornerstone of modern intelligent transportation systems. While the recent advent of Large Language Models (LLMs) has introduced remarkable zero-shot capabilities to urban computing, their immense parametric scale and exorbitant deployment costs fundamentally restrict their viability for real-time, large-scale traffic operations. Concurrently, mainstream spatio-temporal forecasting methods remain heavily tethered to explicit topological graph structures or computationally expensive attention mechanisms, which inherently introduce severe scalability bottlenecks. These limitations underscore a critical divergence in current research: an urgent necessity to develop highly efficient, lightweight predictive architectures capable of bridging the gap between theoretical accuracy and practical deployability.To address this deployment imperative, we propose the Spatio-Temporal Multi-view MLP Network (STMMN), a highly scalable, graph-free computational framework. Diverging from traditional graph-dependent paradigms, STMMN utilizes an adaptive spatial clustering multilayer perceptron (MLP) module to autonomously extract latent spatial homogeneity. This mechanism streamlines spatial dimensionality, effectively circumventing the rigidity of predefined topologies and drastically reducing learning complexity. Furthermore, we introduce a temporal multi-view aggregation module that seamlessly integrates distinct temporal branches, specifically capturing proximity, trend, and periodicity, to dynamically fuse multi-scale temporal dependencies without relying on exhaustive attention maps.Extensive experiments conducted on multiple real-world, large-scale datasets confirm that STMMN achieves state-of-the-art prediction accuracy while exhibiting exceptional computational efficiency. Ultimately, this research validates the immense practical value of lightweight, knowledge-aware models in overcoming the deployment bottlenecks of large-scale urban engineering operations.
The Structure-Guided Spatiotemporal Attention Graph Neural Network is proposed, offering a mechanistic account of the model's decision-making process while ensuring robust forecasting by aligning attention-based reasoning with identified macroscopic dependencies and preventing over-reliance on ephemeral local noise.
The key challenge in traffic flow prediction lies in modeling complex spatio-temporal dependencies effectively. While graph neural networks have shown promise, existing methods face two critical limitations: (1) static graph construction approaches fail to adapt to real-time network dynamics, and (2) prevailing spatio-temporal models neglect both interactive dependency learning and node-specific pattern variations due to spatial heterogeneity. A model based on Interactive Progressive Graph Convolution with Multi-view learning (IPGMVL) is proposed, which introduces three key innovations: First, progressive graph convolution dynamically adjusts edge weights through trend similarity learning, capturing real-time spatial evolution. Second, a multi-view interactive learning mechanism incorporates spatio-temporal heterogeneous patterns for comprehensive dependency modeling. Third, the fast parallel learning (FPL) module is used to realize the synchronous and efficient mining of spatio-temporal features through parameter streamlining, while the serial learning (SL) module expands the serial receptive field and avoid information coverage to further enhance the modeling capability of spatio-temporal dependencies. Experimental results demonstrate IPGMVL's superior performance across four benchmark datasets, establishing new state-of-the-art standards while maintaining computational efficiency. This advancement highlights the importance of dynamic graph adaptation and interactive learning in traffic prediction systems.
This framework introduces an adaptive graph learning module that dynamically infers meaningful connectivity relationships among traffic sensors—not relying on fixed geographic or distance-based assumptions—but instead leveraging real-time traffic correlations and node-level embeddings, enabling effective modeling of both localized spatial interactions and multi-scale temporal dependencies across varying prediction horizons.
Zhengxu Luan, Huan Wang, Miaobowen Wang et al.· Computers and artificial int...· 0 citations
Spatio-temporal graph neural networks achieve strong traffic forecasting accuracy, yet their robustness under out-of-distribution (OOD) conditions, such as traffic incidents, remains poorly understood. We propose SRGNet, a spectrally-regularized graph network combining three targeted innovations: (1) spectral normalization on all weight matrices to bound the Lipschitz constant; (2) disruption-aware training augmentation synthesizing incident-like flow drops; and (3) stochastic depth creating an implicit ensemble. We evaluate on PEMS-BAY using an impact-verified OOD protocol with 996 real incidents ( 30% flow reduction). SRGNet achieves the lowest OOD degradation (+116.0%) among competitive models, the best local OOD RMSE (0.987) at the most-impacted sensors, and a standard RMSE of 0.3123, demonstrating the best accuracy–robustness tradeoff.
Huu Dang Khoi Nguyen, T. Le· E3S Web of Conferences· 0 citations
Experimental results on the public PEMS04 and PEMS08 datasets demonstrate that the proposed ESDG-ALSTM model significantly improves forecasting accuracy, confirming that ESDG-ALSTM is more sensitive to abrupt events and multimodal evolution patterns and can effectively enhance prediction performance in complex traffic flow scenarios.
Guozheng Li, Bai-Jing Wu, Ke Gao et al.· Frontiers of Computer Scienc...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
