Analytic solution of neural network with disordered lateral inhibition

Area of Science:

• Computational Neuroscience
• Statistical Mechanics
• Artificial Neural Networks

Background:

• The replica method is crucial for analyzing disordered systems like the Sherrington-Kirkpatrick (SK) model and neural networks.
• Recurrent neural networks with lateral inhibition are used to model cognitive functions, but can exhibit unstable activity.
• Disorder in neural interactions can significantly impact network dynamics and function.

Purpose of the Study:

• To investigate the effect of disorder in lateral inhibition interactions on recurrent neural network behavior.
• To analytically solve a model with distance-dependent neuronal interactions and disorder.
• To understand the phase transitions and identify conditions for stable network activity.

Main Methods:

• Application of the replica method to analyze disordered recurrent neural networks with lateral inhibition.
• Analytical solution of the model, incorporating distance-dependent interactions.
• Bifurcation analysis to map phase boundaries (paramagnetic, ferromagnetic, spin-glass, localized).

Main Results:

• The study identifies distinct phases, including a localized phase characterized by bump-like activity.
• Disordered interactions were found to stabilize the bump activity, mitigating drift observed in conventional networks.
• The model with distance-dependent interactions was solved analytically, providing insights into phase transitions.

Conclusions:

• Disorder in lateral inhibition interactions can be beneficial, leading to more stable network dynamics.
• The localized phase with stabilized bump activity offers a potential mechanism for spatial working memory and visual cortex columnar activity.
• The replica method provides a powerful analytical tool for studying complex neural network models with disorder.