Surrogate population models for large-scale neural simulations.

Area of Science:

• Computational Neuroscience
• Neural Engineering
• Systems Neuroscience

Background:

• Modeling small brain regions in isolation is unrealistic due to complex interconnections.
• Simulating large-scale neural systems in detail is computationally impractical.
• Existing models struggle to balance biological realism with computational efficiency.

Purpose of the Study:

• To develop a novel multiscale modeling approach for neural systems.
• To construct efficient surrogate models that approximate the aggregate outputs of neuron populations.
• To reduce computational demands for simulating large neural networks.

Main Methods:

• Developed surrogate models for populations of neuron models with correlated activity and specific connections.
• Modeled weighted sums of spikes by interpolating over latent variables in population activity.
• Utilized linear filters operating on Gaussian random variables to approximate spike-related fluctuations, specifically within Neural Engineering Framework (NEF) circuits.

Main Results:

• Surrogate models closely approximate network behavior with orders-of-magnitude reduction in computational demands.
• Approximate spike rasters for specific neurons can be derived from surrogate model simulations.
• Simulations can utilize significantly larger step sizes (e.g., 20 ms) due to the absence of individual spike modeling.

Conclusions:

• The proposed surrogate modeling approach offers a computationally efficient alternative for large-scale neural simulations.
• This method provides a clear relationship between the surrogate and detailed neural models.
• Potential extensions include application to non-NEF networks and more complex neuron models.