Damba-ST: Domain-Adaptive Mamba for Efficient Urban Spatio-Temporal Prediction

Training urban spatio-temporal foundation models that generalize well across diverse regions and cities 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 spatio-temporal prediction. However, directly applying Mamba as a spatio-temporal backbone leads to negative transfer and severe performance degradation. This is primarily due to 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 cross-domain commonalities. Extensive experiments demonstrate the generalization and efficiency of Damba-ST. It achieves state-of-the-art performance on prediction tasks and demonstrates strong zero-shot generalization, enabling seamless deployment in new urban environments without extensive retraining or fine-tuning.

💡 Research Summary

**
The paper tackles two fundamental challenges in urban spatio‑temporal forecasting: (1) the severe heterogeneity of data across cities and regions, which leads to negative transfer when a single model is trained on fused datasets, and (2) the quadratic computational and memory cost of Transformer‑based foundations that hampers scalability to large traffic graphs and long horizons. To address these issues, the authors build on Mamba, a state‑space model (SSM) that achieves linear‑time complexity through selective parameterisation and parallel associative scanning. While Mamba offers efficiency, its original design for language sequences lacks mechanisms to handle cross‑domain distribution shifts, and its recursive hidden‑state updates can amplify domain‑specific patterns, degrading zero‑shot generalisation.

Damba‑ST (Domain‑Adaptive Mamba for Spatio‑Temporal prediction) introduces a Domain‑Adaptive State Space Model (DASSM) that explicitly partitions the latent representation into a shared subspace C (capturing universal temporal regularities such as daily/weekly cycles) and domain‑specific subspaces Sᵢ (encoding each city’s unique spatial layout, event patterns, etc.). To bridge the gap between domains, three Domain Adapters are added as learnable embeddings. These adapters act as proxies, analogous to the