Exploiting noise as a resource for computation and learning in spiking neural networks

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Spiking neural networks (SNNs) are crucial for brain-inspired AI.
• Current SNN models often ignore the brain's inherent neural noise.
• Deterministic models limit the understanding of neural computation.

Purpose of the Study:

• Introduce a novel noisy SNN (NSNN) framework.
• Develop a noise-driven learning (NDL) rule.
• Exploit noisy neural dynamics for enhanced computation.

Main Methods:

• Incorporated noisy neuronal dynamics into SNNs.
• Developed and applied the noise-driven learning (NDL) rule.
• Evaluated NSNN performance against deterministic SNNs.

Main Results:

• NSNNs demonstrate scalable, flexible, and reliable computation and learning.
• Achieved competitive performance and improved robustness against perturbations.
• Showed better reproduction of probabilistic computation in neural coding.

Conclusions:

• NSNNs offer a powerful tool for machine learning and neuromorphic intelligence.
• The NDL rule enhances SNN capabilities by embracing neural noise.
• This framework advances computational neuroscience research by modeling realistic neural dynamics.