Related Experiment Video
Updated: Jan 15, 2026
09:44
Author Spotlight: Advancing Large-Scale Neural Dynamics Through HD-MEA Technology
Published on: March 8, 2024
5.8K
Adaptive dendritic plasticity in brain-inspired dynamic neural networks for enhanced multi-timescale feature
Jiayi Mao1, Hanle Zheng1, Huifeng Yin1
1Center for Brain Inspired Computing Research, Department of Precision Instrument, Tsinghua University, Beijing, 100084, China.
Summary
This study introduces a dynamic Spiking Neural Network (SNN) with enhanced dendritic heterogeneity to improve multi-timescale temporal feature extraction. The novel approach boosts performance in processing complex temporal signals.
Area of Science:
- Computational Neuroscience
- Artificial Intelligence
Background:
- Brain-inspired neural networks, particularly Spiking Neural Networks (SNNs), show promise for temporal processing.
- Traditional SNNs face challenges in capturing multi-timescale features due to static parameters and low-precision spikes.
Purpose of the Study:
- To enhance the multi-timescale feature extraction capability of SNNs.
- To address limitations in current SNN models for complex temporal signal processing.
Main Methods:
- Proposed a dynamic SNN incorporating a Leaky Integrate Modulation neuron with Dendritic Heterogeneity (DH-LIM).
- Introduced an Adaptive Dendritic Plasticity (ADP) mechanism for dynamic adjustment of dendritic timing factors.
- Replaced traditional spike activities with a continuous modulation mechanism for nonlinear behavior preservation.
Main Results:
- The proposed DH-LIM neuron model enhances feature expression and preserves nonlinear behaviors.
- The ADP mechanism enables capture of both rapid and slow temporal patterns by adapting to input signal frequencies.
- Extensive experiments demonstrated excellent performance on datasets with rich temporal features.
Conclusions:
- The developed dynamic SNN with enhanced dendritic heterogeneity offers improved multi-timescale feature extraction.
- This approach provides novel solutions for optimizing SNNs, highlighting broad application potential in temporal signal processing.
