Search papers, labs, and topics across Lattice.
This paper introduces a novel method called Temporal-Spectral Alignment with Frequency Adaptation (SAFA) for source-free domain adaptation of time-series data, addressing both feature shifts and temporal drifts. By modeling the source domain at multiple scales and incorporating a trainable frequency adaptation module, SAFA effectively aligns target signals in the frequency domain with the source distribution. Experimental results across various benchmark datasets show that SAFA significantly outperforms existing methods in robustness and adaptation efficacy.
Frequency adaptation can dramatically enhance source-free time-series domain adaptation, aligning target signals with source distributions more effectively than previous methods.
The goal of source-free domain adaptation (SFDA) for time-series data is to transfer knowledge from a pre-trained source model to an unlabeled target domain without requiring access to source data, while addressing feature shift and temporal drift inherent in the signals. Although existing approaches have explored temporal dynamics in unsupervised source-free adaptation, they largely overlook spectral shifts in time-series data. Towards this end, we propose a novel approach termed temporal-Spectral Alignment with Frequency Adaptation (SAFA) for source-free time-series domain adaptation. Specifically, we first model the source domain at multiple scales by jointly capturing temporal dependencies and spectral characteristics. To adapt time-series data in the target domain, we introduce a trainable frequency adaptation module that modulates the phase and amplitude of target signals in the frequency domain to align them with the source distribution. Extensive experiments on multiple benchmark datasets demonstrate the efficacy and robustness of SAFA.
