Sparse Coding and Temporal Pattern Learning Co-Mediated by Dual Spike-Timing-Dependent Plasticity in a Multilayer...

Area of Science:

• Computational neuroscience
• Neural networks
• Synaptic plasticity

Background:

• Excitatory-inhibitory (E-I) circuits are crucial for neural computation.
• Understanding multilayer learning dynamics in E-I circuits is limited.
• Spiking neural networks (SNNs) offer a biologically plausible model for neural processing.

Purpose of the Study:

• Investigate cooperative learning in multilayer feedforward SNNs with E-I connectivity.
• Examine the roles of excitatory and inhibitory spike-timing-dependent plasticity (eSTDP and iSTDP).
• Analyze the impact of E-I interactions on neural coding, sparse coding, and temporal pattern learning.

Main Methods:

• Constructed a multilayer feedforward SNN using Izhikevich neurons (regular spiking and fast spiking).
• Implemented intra-layer E-I connectivity and simulated eSTDP and iSTDP.
• Performed parameter grid scans and analyzed network outputs for sparseness and temporal pattern learning.

Main Results:

• Fast spiking (FS) cell-mediated inhibition counteracts firing rate adaptation, promoting weight differentiation.
• iSTDP expands stable learning regions and co-evolves with eSTDP.
• The network demonstrates enhanced sparse coding and improved temporal pattern selectivity (d' ≈ 1.90x without FS inhibition).
• FS circuits maintain higher stimulus information in deeper layers.

Conclusions:

• Cortical E-I cooperative plasticity enhances neural coding and learning.
• Adaptive inhibitory regulation is key for neuromorphic system design.
• The model provides computational evidence for the functional role of E-I circuits in the brain.