TS-LIF

March 10, 2025 ยท View on GitHub

[ICLR 2025] Official Implementation of "TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting"

Installation

To install TS-LIF within (SeqSNN) in a new conda environment:

conda create -n SeqSNN python=[3.8, 3.9, 3.10]
conda activate SeqSNN
git clone https://github.com/microsoft/SeqSNN/
cd SeqSNN
pip install .

If you would like to make changes and run your experiments, use:

pip install -e .

Training

Take the TS-GRU model as an example:

python entry.tsforecast.py /path/exp/forecast/spikegru/spikegru_electricity.yml"

You can change the yml configuration files as you want.

Datasets

Metr-la and Pems-bay datasets are available on Google Drive and Baidu Yun.
Solar and Electricity datasets can be downloaded from GitHub.

Acknowledgement

This repo is built upon (Lv's Repo) [https://github.com/microsoft/SeqSNN], and we show sincerely thanks for @Changze Lv's initial contribution.

Citing

If our work has inspired your research, you can consider citing us:

@inproceedings{shibots,
  title={TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting},
  author={SHIBO, FENG and Feng, Wanjin and Gao, Xingyu and Zhao, Peilin and Shen, Zhiqi},
  booktitle={The Thirteenth International Conference on Learning Representations}
}

@article{lv2024efficient,
  title={Efficient and effective time-series forecasting with spiking neural networks},
  author={Lv, Changze and Wang, Yansen and Han, Dongqi and Zheng, Xiaoqing and Huang, Xuanjing and Li, Dongsheng},
  journal={arXiv preprint arXiv:2402.01533},
  year={2024}
}