A Synchronization-Driven Learning Rule for Pattern Separation in Self-Organizing Probabilistic Spiking Neural...

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Neuroscience-inspired neural networks offer a path to integrate biological principles with AI.
• Spiking neural networks (SNNs) are computational models that mimic biological neurons.

Purpose of the Study:

• To propose and evaluate a novel synchronization-based synaptic learning rule for self-organizing probabilistic spiking neural networks (PSNNs).
• To investigate the role of feedback inhibition in network dynamics, stability, and pattern separation.
• To compare the proposed learning rule with conventional Hebbian learning.

Main Methods:

• Developed a novel synchronization-based synaptic learning rule for PSNNs with feedback inhibition.
• Systematically analyzed the influence of feedback inhibition on network dynamics, stability, synchronization, and pattern separation.
• Conducted comparative analysis against Hebbian learning.
• Embedded the trained network in a simulated autonomous agent for a navigation task.

Main Results:

• Moderate feedback inhibition optimized the balance between excitation and inhibition, maximizing pattern separation.
• The proposed synchronization-based learning rule demonstrated superior efficiency and stability in pattern separation compared to Hebbian learning.
• The trained network successfully navigated a simulated environment, identifying and avoiding learned obstacle patterns.

Conclusions:

• Feedback inhibition and synchronization-driven plasticity are crucial for self-organizing spiking systems.
• Biologically inspired learning mechanisms hold significant potential for computational neuroscience, neuromorphic computing, and cognitive robotics.