Freezing chaos without synaptic plasticity.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Dynamical Systems

Background:

• High-dimensional neural dynamics often exhibit ubiquitous chaos.
• Chaotic fluctuations can impair neural information processing.
• Synaptic plasticity, like Hebbian learning, is a traditional method for stabilizing neural dynamics.

Purpose of the Study:

• To introduce a novel method for stabilizing neural dynamics without relying on synaptic plasticity.
• To investigate the effects of an Onsager reaction term on recurrent neural dynamics.
• To explore the potential of gradient dynamics for computational neuroscience tasks.

Main Methods:

• Incorporating an Onsager reaction term into standard recurrent neural network dynamics.
• Formulating the neural dynamics as a gradient system.
• Analyzing the behavior of these gradient dynamics in simplified and biologically realistic neural networks.

Main Results:

• The addition of the Onsager term transforms the dynamics into a gradient form.
• This gradient formulation allows for the stabilization of chaotic fluctuations by reducing kinetic energy.
• The freezing effect was demonstrated in networks with both excitatory and inhibitory neurons.
• The gradient dynamics proved effective for computational tasks like stimulus recall and prediction.

Conclusions:

• A novel, plasticity-free method using gradient dynamics can effectively stabilize chaotic neural activity.
• This approach offers an alternative to traditional synaptic plasticity for managing neural noise.
• The developed gradient dynamics show promise for enhancing neural computation and information processing.