Synchronization and Inter-Layer Interactions of Noise-Driven Neural Networks

Area of Science:

• Computational Neuroscience
• Artificial Neural Networks

Background:

• Neural networks exhibit complex dynamics, transitioning between asynchronous and synchronous states.
• Understanding these transitions is crucial for modeling brain function and developing AI.

Purpose of the Study:

• To investigate the phase diagram of single-layer and two-layer neural networks.
• To analyze the transition from background activity state (BAS) to synchronous firing state (SFS) under noise and learning.
• To examine the influence of inter-layer interactions on this transition and repair mechanism.

Main Methods:

• Utilized the Hodgkin-Huxley (HH) model for simulating neuronal activity.
• Employed spike-timing-dependent plasticity (STDP) and inverse STDP learning rules.
• Varied network connectivity, learning efficacy, neuron positioning (random, grid, lognormal), and inter-layer connections (random, preferential).

Main Results:

• Demonstrated the transition from BAS to SFS by adjusting network connectivity and learning efficacy.
• Showcased how a synchronous firing state (SFS) layer can induce synchronous dynamics in a background activity state (BAS) layer.
• Identified optimal conditions for a BAS to SFS repair mechanism.

Conclusions:

• The transition to synchronous firing is controllable via network parameters and learning.
• Inter-layer interactions play a significant role in modulating neural network dynamics.
• Lognormal neuron positioning with preferential inter-layer connections most effectively facilitates the BAS to SFS repair mechanism.