Evolutionary algorithm optimization of biological learning parameters in a biomimetic neuroprosthesis.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Biomimetic simulations enhance understanding of complex brain dynamics.
• Closed-loop neuroprostheses integrating simulations offer therapeutic potential for neurological and psychiatric disorders.
• Dynamic data-driven systems are crucial for real-time adaptation in neuroprosthetics.

Purpose of the Study:

• To develop and optimize a biomimetic neuroprosthesis capable of real-time adaptation.
• To employ biological learning rules, specifically spike-timing dependent plasticity and reinforcement learning, for parameter optimization.
• To train a cortical spiking model for controlling a virtual musculoskeletal arm based on macaque brain activity.

Main Methods:

• Utilized spike-timing dependent plasticity and reinforcement learning instead of traditional machine learning for synaptic weight adaptation.
• Employed parallel evolutionary algorithms with an island model approach for efficient metaparameter optimization.
• Trained a cortical spiking model using macaque brain activity data to drive a biomechanically realistic virtual arm.

Main Results:

• The optimized biomimetic neuroprosthesis system successfully reproduced macaque experimental motor task data.
• Demonstrated the ability to link realistic neuronal dynamics to behavioral output.
• Validated the efficiency of evolutionary algorithms in tuning parameters for multiscale systems.

Conclusions:

• The developed techniques efficiently tune parameters in multiscale systems, bridging neuronal dynamics and behavior.
• This approach offers a valuable tool for advancing neuroscience research and neuroprosthetic applications.
• Biomimetic neuroprostheses show promise for ameliorating motor, psychiatric, and memory-related brain disorders.