Related Experiment Video
Updated: Jul 3, 2026
07:34
A Simple Stimulatory Device for Evoking Point-like Tactile Stimuli: A Searchlight for LFP to Spike Transitions
Published on: March 25, 2014
9.9K
Methodology based on spiking neural networks for univariate time-series forecasting
1Department of Automatic Control and Systems Engineering, Faculty of Engineering of Bilbao, University of the Basque Country (UPV/EHU), Plaza Ingeniero Torres Quevedo, 1, Bilbao, 48013, Basque Country, Spain.
Summary
This study introduces a new training method for Spiking Neural Networks (SNNs) for time-series forecasting. The approach achieves robust and energy-efficient predictions across various datasets.
Area of Science:
- Artificial Intelligence
- Computational Neuroscience
- Machine Learning
Background:
- Spiking Neural Networks (SNNs) excel at processing spatiotemporal data with low energy use.
- Existing SNN applications predominantly focus on classification, with fewer on time-series forecasting.
Purpose of the Study:
- To present a novel, general training methodology for univariate time-series forecasting using SNNs.
- To address the gap in SNN applications for forecasting problems.
Main Methods:
- The methodology employs a Pulse-Width Modulation (PWM) based encoding-decoding algorithm.
- Supervised training is achieved using a Surrogate Gradient method for one-step-ahead forecasting.
- Validation is performed on diverse datasets, including sine-wave, UCI, and real-world data.
Main Results:
- The proposed SNN methodology yields highly satisfactory forecasting results, with Mean Absolute Error (MAE) ranging from 0.0094 to 0.2891.
- Effective forecasting performance is demonstrated across datasets with varying characteristics and application domains.
- Robust results are achieved even with a single weight initialization, highlighting computational and energy efficiency.
Conclusions:
- The developed SNN training methodology offers a versatile and effective solution for univariate time-series forecasting.
- The approach leverages the inherent low-power and computational advantages of SNNs for forecasting tasks.
- This work expands the applicability of SNNs to forecasting, demonstrating their potential beyond classification.
