Real-time Neural Connectivity Inference with Presynaptic Spike-driven Spike Timing-Dependent Plasticity

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computer Engineering

Background:

• Understanding neuronal circuitry connectivity is key for efficient brain-like artificial intelligence.
• Previous neural connectivity inference methods using 2D memristor arrays face challenges in large-scale implementation due to device variations and online parameter adaptation needs.

Purpose of the Study:

• To propose a novel neural connectivity inference method for large-scale neuromorphic systems.
• To utilize one-dimensional spiking neurons with specific plasticity learning rules for efficient inference.

Main Methods:

• Developed a neural connectivity inference method employing one-dimensional spiking neurons.
• Implemented spike timing-dependent plasticity (STDP) and presynaptic spike-driven STDP learning rules.
• Simulated 12 ground-truth neural networks (1D, 8 and 64 neurons) and analyzed inference accuracy using Matthews correlation coefficient, sensitivity, and specificity.

Main Results:

• The proposed learning process effectively reduces the number of required spiking neurons by 50%.
• High correlation was observed between ground-truth neural connectivity and the inferred spiking neural networks.
• Validation demonstrated the robustness and potential of the presynaptic spike-driven STDP rule for large-scale inference.

Conclusions:

• The proposed method offers an efficient approach for neural connectivity inference in large-scale neuromorphic systems.
• The use of 1D spiking neurons and specific plasticity rules facilitates hardware realization.
• This research contributes to the advancement of brain-like artificial intelligence and neuromorphic computing.