Synaptic weighting for physiological responses in recurrent spiking neural networks

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neural Network Modeling

Background:

• Recurrent neural networks (RNNs) model neuro-physiological phenomena like sensorimotor integration.
• Specifying synaptic weights in RNNs for desired neuron spiking responses is challenging due to cyclic connections.

Purpose of the Study:

• To decouple connection topology from neuron response characteristics in RNNs.
• To enable a priori specification of synaptic weights for targeted neural responses.
• To advance computational models of neuro-physiology.

Main Methods:

• Utilized a mean field assumption for large neuronal populations with uncorrelated synaptic inputs.
• Developed a method to specify neuron steady-state responses independently of network topology.
• Validated the synaptic weighting approach through two case studies.

Main Results:

• Demonstrated that connection topology and neuron response characteristics can be decoupled in RNNs.
• Showed that steady-state responses can be specified irrespective of the underlying connection structure.
• Provided empirical validation for the proposed synaptic weighting strategy.

Conclusions:

• The developed method allows for independent control over neural network structure and function.
• This approach facilitates the creation of more accurate and predictable computational models of the brain.
• Offers a novel way to engineer neural network dynamics for specific neuro-physiological applications.