[KDD 2025] Temporal Restoration and Spatial Rewiring for Source-Free Multivariate Time Series Domain Adaptation

by: Peiliang Gong, Yucheng Wang, Min Wu, Zhenghua Chen, Xiaoli Li, Daoqiang Zhang

Accepted in the 31st SIGKDD Conference on Knowledge Discovery and Data Mining - Research Track.

Requirements:

einops==0.7.0
matplotlib==3.5.0
pandas==2.2.1
scipy==1.12.0
seaborn==0.13.2
torch==2.1.0
torchaudio==2.1.0
torchdata==0.7.1
torchmetrics==1.2.1
torchvision==0.16.0
wandb==0.16.5

Datasets:

We used three public datasets in this study.

• UCIHAR
• SSC
• WISDM

Training procedure:

• Here, we provide a demo for running the experiments.
  To train a model using the following script file:

python trainers/train.py --run_description demo --da_method TERSE --dataset HAR --backbone TemporalSpatialNN_new --num_runs 3

Results:

At the end of all runs, the overall average and standard deviation results will be saved in the save_dir directory.

Citation

If you found this work useful for you, please consider citing it.

@inproceedings{terse,
  author = {Gong, Peiliang and Wang, Yucheng and Wu, Min and Chen, Zhenghua and Li, Xiaoli and Zhang, Daoqiang},
  title = {Temporal Restoration and Spatial Rewiring for Source-Free Multivariate Time Series Domain Adaptation},
  booktitle={31st SIGKDD Conference on Knowledge Discovery and Data Mining - Research Track},
  year = {2025}
}