Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting

Shibo Feng, Wanjin Feng, Xingyu Gao, Peilin Zhao, Zhiqi Shen

TL;DR

The paper addresses the challenge that traditional LIF-based SNNs struggle to capture long-range temporal dependencies and multi-scale dynamics in time-series forecasting. It introduces the Temporal Segment LIF (TS-LIF), a dual-compartment neuron with dendritic and somatic pathways, plus direct somatic current injection and dendritic spikes, enabling explicit multi-scale temporal processing. The authors provide stability and frequency-response analyses to justify robustness and derive transfer functions that reveal compartment-specific frequency separation. Empirical results across four datasets and multiple backbones show consistent improvements over LIF-based baselines and good robustness to missing data, highlighting TS-LIF’s practical value for energy-efficient, accurate forecasting in resource-constrained settings.

Abstract

Spiking Neural Networks (SNNs) offer a promising, biologically inspired approach for processing spatiotemporal data, particularly for time series forecasting. However, conventional neuron models like the Leaky Integrate-and-Fire (LIF) struggle to capture long-term dependencies and effectively process multi-scale temporal dynamics. To overcome these limitations, we introduce the Temporal Segment Leaky Integrate-and-Fire (TS-LIF) model, featuring a novel dual-compartment architecture. The dendritic and somatic compartments specialize in capturing distinct frequency components, providing functional heterogeneity that enhances the neuron's ability to process both low- and high-frequency information. Furthermore, the newly introduced direct somatic current injection reduces information loss during intra-neuronal transmission, while dendritic spike generation improves multi-scale information extraction. We provide a theoretical stability analysis of the TS-LIF model and explain how each compartment contributes to distinct frequency response characteristics. Experimental results show that TS-LIF outperforms traditional SNNs in time series forecasting, demonstrating better accuracy and robustness, even with missing data. TS-LIF advances the application of SNNs in time-series forecasting, providing a biologically inspired approach that captures complex temporal dynamics and offers potential for practical implementation in diverse forecasting scenarios. The source code is available at https://github.com/kkking-kk/TS-LIF.

TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting

TL;DR

The paper addresses the challenge that traditional LIF-based SNNs struggle to capture long-range temporal dependencies and multi-scale dynamics in time-series forecasting. It introduces the Temporal Segment LIF (TS-LIF), a dual-compartment neuron with dendritic and somatic pathways, plus direct somatic current injection and dendritic spikes, enabling explicit multi-scale temporal processing. The authors provide stability and frequency-response analyses to justify robustness and derive transfer functions that reveal compartment-specific frequency separation. Empirical results across four datasets and multiple backbones show consistent improvements over LIF-based baselines and good robustness to missing data, highlighting TS-LIF’s practical value for energy-efficient, accurate forecasting in resource-constrained settings.

Abstract

Spiking Neural Networks (SNNs) offer a promising, biologically inspired approach for processing spatiotemporal data, particularly for time series forecasting. However, conventional neuron models like the Leaky Integrate-and-Fire (LIF) struggle to capture long-term dependencies and effectively process multi-scale temporal dynamics. To overcome these limitations, we introduce the Temporal Segment Leaky Integrate-and-Fire (TS-LIF) model, featuring a novel dual-compartment architecture. The dendritic and somatic compartments specialize in capturing distinct frequency components, providing functional heterogeneity that enhances the neuron's ability to process both low- and high-frequency information. Furthermore, the newly introduced direct somatic current injection reduces information loss during intra-neuronal transmission, while dendritic spike generation improves multi-scale information extraction. We provide a theoretical stability analysis of the TS-LIF model and explain how each compartment contributes to distinct frequency response characteristics. Experimental results show that TS-LIF outperforms traditional SNNs in time series forecasting, demonstrating better accuracy and robustness, even with missing data. TS-LIF advances the application of SNNs in time-series forecasting, providing a biologically inspired approach that captures complex temporal dynamics and offers potential for practical implementation in diverse forecasting scenarios. The source code is available at https://github.com/kkking-kk/TS-LIF.

Paper Structure

This paper contains 36 sections, 1 theorem, 27 equations, 7 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Modeling Long-Term Dependencies in SNNs
  4. Time Series Forecasting
  5. Preliminaries
  6. Time Series Problem Setting
  7. LIF Neuron Model
  8. Dendrites and Soma
  9. Methodology
  10. Temporal Segment LIF Neuron
  11. Stability Analysis
  12. Frequency Response Analysis
  13. Example: Frequency Separation in Practice
  14. Experiments
  15. Temporal Analysis
  16. ...and 21 more sections

Key Result

Theorem 1

The system governed by the following dynamics: has eigenvalues: For the system to remain stable, it is necessary that $|\lambda| < 1$ for both eigenvalues.

Figures (7)

  • Figure 1: Diagram of Neuronal Signal Processing and Integration: (a) Structural organization of neuronal signal transmission, highlighting axosomatic and axodendritic synapses. (b) A generalized two-compartment spiking neuron model, applicable to TC-LIF or LM-H models, with dendritic (gray) and somatic (orange) compartments. (c) Proposed TS-LIF model with the newly introduced direct somatic current injection and dendritic spike generation (highlighted in red). (d) Time series decomposition and spike output generation in the TS-LIF model.
  • Figure 2: (a) TC-LIF Model Response
  • Figure 3: (b) TS-LIF Model Response
  • Figure 5: Impact of Training Timesteps on Forecasting Performance of TS-LIF Model Across Different Architectures. Solar dataset with a prediction length of 24 and Electricity dataset with a prediction length of 96. We plot mean and std for each experiment over 3 different random seeds.
  • Figure 6: Forecasting Accuracy Comparison of TS-LIF and TC-LIF Neurons on Solar and Electricity datasets with a prediction length of 24.
  • ...and 2 more figures

Theorems & Definitions (2)

  • Theorem 1
  • proof