Stabilization of Hebbian neural nets by inhibitory learning

Area of Science:

• Neuroscience
• Computational Neuroscience
• Neural Modeling

Background:

• Hebbian learning models describe synaptic reinforcement based on simultaneous pre- and post-synaptic neuron activity.
• Previous models required non-linear normalization or cut-off conditions to prevent synaptic growth instability.
• Recurrent inhibition via modifiable inhibitory synapses was not fully integrated into stabilizing these models.

Purpose of the Study:

• To develop a stabilized Hebbian neural model.
• To investigate the role of recurrent inhibition in neural network stability.
• To analyze the response characteristics and emergent properties of the stabilized model.

Main Methods:

• Introducing recurrent inhibition with modifiable inhibitory synapses into the Hebbian model.
• Formulating linearized equations that are tensor invariant under state space rotations.
• Analyzing the system's response to stimulation using independent modes of activity.

Main Results:

• The inclusion of recurrent inhibition stabilizes the neural network, removing the need for non-linear normalization.
• The model exhibits a response-linear behavior under slow synaptic changes and maintains self-consistency.
• The linearized equations allow derivation of stimulation responses as independent activity modes, identifiable with cell assemblies.
• An infinite set of equivalent solutions was found to exist.

Conclusions:

• Recurrent inhibition is a key mechanism for stabilizing Hebbian neural networks.
• The proposed model offers a more consistent and linear framework for understanding neural dynamics and cell assembly formation.
• The findings provide a theoretical basis for analyzing neural network responses and emergent structures.