TY - GEN
T1 - Damba-ST
T2 - 42nd International Conference on Data Engineering, ICDE 2026
AU - An, Rui
AU - Zhang, Yifeng
AU - Liang, Ziran
AU - Fan, Wenqi
AU - Liang, Yuxuan
AU - Shang, Xuequn
AU - Li, Qing
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Training urban spatio-temporal foundation models that generalize well across diverse regions and cities (i.e., crossdomain scenarios) is critical for deploying urban services in unseen or data-scarce regions. Recent studies have typically focused on fusing cross-domain spatio-temporal data to train unified Transformer-based models. However, these models suffer from quadratic computational complexity and high memory overhead, limiting their scalability and practical deployment. Inspired by the efficiency of Mamba, a state-space model with linear time complexity, we explore its potential for efficient urban spatiotemporal prediction. However, directly applying Mamba as a spatio-temporal backbone leads to negative transfer and severe performance degradation in unseen regions. This is primarily due to inherent spatio-temporal heterogeneity and the recursive mechanism of Mamba's hidden state updates, which limit cross-domain generalization. To overcome these challenges, we propose Damba-ST, a novel domain-adaptive Mamba-based model for efficient urban spatio-temporal prediction. Damba-ST retains Mamba's linear complexity advantage while significantly enhancing its adaptability to heterogeneous domains. Specifically, we introduce two core innovations: (1) a Domain-Adaptive State Space Model that partitions the latent representation space into a shared subspace for learning cross-domain commonalities and independent, domain-specific subspaces for capturing intra-domain discriminative features; (2) three distinct Domain Adapters, which serve as domain-aware proxies to bridge disparate domain distributions and facilitate the alignment of crossdomain commonalities. Extensive experiments demonstrate the generalization and efficiency of Damba-ST. It achieves state-ofthe-art performance on prediction tasks and demonstrates strong zero-shot generalization, enabling seamless deployment in new urban environments without extensive retraining or fine-tuning. The code is available at https://github.com/RuiAN77/Damba-ST.
AB - Training urban spatio-temporal foundation models that generalize well across diverse regions and cities (i.e., crossdomain scenarios) is critical for deploying urban services in unseen or data-scarce regions. Recent studies have typically focused on fusing cross-domain spatio-temporal data to train unified Transformer-based models. However, these models suffer from quadratic computational complexity and high memory overhead, limiting their scalability and practical deployment. Inspired by the efficiency of Mamba, a state-space model with linear time complexity, we explore its potential for efficient urban spatiotemporal prediction. However, directly applying Mamba as a spatio-temporal backbone leads to negative transfer and severe performance degradation in unseen regions. This is primarily due to inherent spatio-temporal heterogeneity and the recursive mechanism of Mamba's hidden state updates, which limit cross-domain generalization. To overcome these challenges, we propose Damba-ST, a novel domain-adaptive Mamba-based model for efficient urban spatio-temporal prediction. Damba-ST retains Mamba's linear complexity advantage while significantly enhancing its adaptability to heterogeneous domains. Specifically, we introduce two core innovations: (1) a Domain-Adaptive State Space Model that partitions the latent representation space into a shared subspace for learning cross-domain commonalities and independent, domain-specific subspaces for capturing intra-domain discriminative features; (2) three distinct Domain Adapters, which serve as domain-aware proxies to bridge disparate domain distributions and facilitate the alignment of crossdomain commonalities. Extensive experiments demonstrate the generalization and efficiency of Damba-ST. It achieves state-ofthe-art performance on prediction tasks and demonstrates strong zero-shot generalization, enabling seamless deployment in new urban environments without extensive retraining or fine-tuning. The code is available at https://github.com/RuiAN77/Damba-ST.
KW - Heterogenous Data Management
KW - Mamba
KW - Spatio-Temporal Prediction
KW - State Space Model (SSM)
UR - https://www.scopus.com/pages/publications/105048987325
U2 - 10.1109/ICDE65706.2026.00105
DO - 10.1109/ICDE65706.2026.00105
M3 - 会议稿件
AN - SCOPUS:105048987325
T3 - Proceedings - 2026 IEEE 42nd International Conference on Data Engineering, ICDE 2026
SP - 1349
EP - 1362
BT - Proceedings - 2026 IEEE 42nd International Conference on Data Engineering, ICDE 2026
A2 - Daudjee, Khuzaima
A2 - Luo, Qiong
A2 - Simitsis, Alkis
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 4 May 2026
ER -