Related Experiment Video
Updated: Jul 12, 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
10.0K
A Parallel Spiking Neural Network Based on Adaptive Lateral Inhibition Mechanism for Objective Recognition
1College of Computer Science and Technology, Harbin Engineering University, Harbin 150001, China.
Computational Intelligence and Neuroscience
|October 24, 2022
Summary
This study introduces a novel Spiking Neural Network (SNN) with adaptive lateral inhibition and dynamic time constants, enhancing pattern recognition. The developed SNN achieves competitive accuracy on static data and surpasses Deep Neural Networks (DNNs) on dynamic and neuromorphic datasets.
Area of Science:
- Machine Learning
- Computational Neuroscience
- Artificial Intelligence
Background:
- Spiking Neural Networks (SNNs) offer biological interpretability and low power consumption but lag behind Deep Neural Networks (DNNs) in accuracy.
- The non-differentiability of spiking neurons hinders direct gradient-based training, and a unified training algorithm for SNNs is lacking.
Purpose of the Study:
- To propose a novel Spiking Neural Network (SNN) architecture inspired by the biological vision system.
- To enhance SNN training by incorporating an adaptive lateral inhibition mechanism and dynamically evolving neuron time constants.
- To evaluate the proposed SNN's effectiveness on static, neuromorphic, and medical imaging datasets, specifically for breast tumor recognition.
Main Methods:
- Development of a parallel convolution Spiking Neural Network (SNN) structure.
- Integration of an adaptive lateral inhibition mechanism to improve neuron interaction.
- Introduction of a method for dynamically evolving neuron time constants during SNN training.
- Validation on static and neuromorphic datasets, and application to breast tumor image recognition.
Main Results:
- The proposed SNN demonstrates significant advantages on dynamic datasets.
- SNN performance on static datasets closely approximates that of Deep Neural Networks (DNNs).
- The SNN architecture surpasses DNN performance on neuromorphic datasets.
- The method shows promise for edge-based tasks like breast tumor recognition, crucial for accurate disease diagnosis.
Conclusions:
- The proposed Spiking Neural Network (SNN) architecture effectively addresses training limitations of traditional SNNs.
- The adaptive lateral inhibition and dynamic time constant evolution contribute to improved neuron diversity and network performance.
- The SNN shows strong potential for applications in pattern recognition, particularly on dynamic and neuromorphic data, and in medical image analysis.
