Hierarchical winner-take-all particle swarm optimization social network for neural model fitting

Area of Science:

• Computational Neuroscience
• Machine Learning
• Optimization Algorithms

Background:

• Particle Swarm Optimization (PSO) is a widely used metaheuristic for optimization and machine learning.
• The social network topology significantly impacts PSO performance.
• Existing PSO networks may not be optimal for high-dimensional or complex inverse problems.

Purpose of the Study:

• Introduce a novel PSO social network topology inspired by winner-take-all (WTA) coding in visual cortical neurons.
• Apply this WTA-PSO variant to the inverse problem of selecting input parameters from auditory neuron tuning curves.
• Evaluate the performance of WTA-PSO against other topologies.

Main Methods:

• Developed a new PSO social network topology based on WTA principles.
• Applied the WTA-PSO algorithm to solve the inverse problem of parameter selection for auditory neuron tuning curves.
• Compared the iteration count and performance of WTA-PSO with existing PSO topologies on problems with varying dimensions.

Main Results:

• The WTA-PSO network demonstrated superior performance on optimization problems exceeding 5 dimensions.
• WTA-PSO achieved optimization at a lower iteration count compared to other PSO topologies.
• The proposed PSO variant successfully recreated auditory frequency tuning curves and modulation transfer functions.

Conclusions:

• The novel WTA-PSO social network topology offers improved efficiency and effectiveness for high-dimensional optimization problems.
• This PSO variant shows significant potential as a tool for computational neuroscience, particularly in modeling auditory neuron responses.
• The findings suggest WTA-inspired topologies can enhance PSO's applicability in complex scientific modeling.