Related Experiment Video
Updated: Jun 15, 2025
11:18
Closed-loop Neuro-robotic Experiments to Test Computational Properties of Neuronal Networks
Published on: March 2, 2015
10.3K
Random heterogeneous spiking neural network for adversarial defense.
Jihang Wang1,2,3,4, Dongcheng Zhao1,3,4,5, Chengcheng Du1,6,3,4
1Brain-inspired Cognitive AI Lab, Institute of Automation, Chinese Academy of Sciences, Beijing, China.
Iscience
|June 13, 2025
Summary
Random Heterogeneous Spiking Neural Networks (RandHet-SNN) enhance robustness against adversarial attacks by introducing neuron diversity. This approach boosts security for spiking neural networks without sacrificing performance.
Area of Science:
- Computational Neuroscience
- Artificial Intelligence
- Machine Learning Security
Background:
- Spiking neural networks (SNNs) mimic biological neurons but are vulnerable to adversarial attacks, similar to artificial neural networks (ANNs).
- Ensuring the robustness of SNNs is crucial for their reliable deployment in real-world applications.
Purpose of the Study:
- To develop a novel SNN architecture, the Random Heterogeneous Spiking Neural Network (RandHet-SNN), to improve defense against adversarial examples.
- To leverage biological neural system characteristics like heterogeneity and stochasticity for enhanced network security.
Main Methods:
- Introduced neuron-level diversity in SNNs by incorporating randomized time decay constants.
- Enabled unique temporal properties for each neuron during every forward pass.
- Evaluated RandHet-SNN performance against various adversarial attacks.
Main Results:
- RandHet-SNN demonstrated significant enhancement in network robustness against adversarial attacks.
- The proposed architecture maintained minimal impact on the network's clean accuracy.
- Results highlight the effectiveness of neuron diversity in bolstering SNN security.
Conclusions:
- RandHet-SNN offers a promising solution for creating robust and energy-efficient SNNs in adversarial conditions.
- The study validates the potential of biologically inspired heterogeneity for improving AI security.
