Divide-and-conquer approach for brain machine interfaces: nonlinear mixture of competitive linear models.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-machine interfaces (BMIs) are crucial for restoring function after neurological injury.
• Accurate decoding of neural signals for movement prediction remains a challenge.
• Existing models often require significant computational resources.

Purpose of the Study:

• To propose a novel divide-and-conquer strategy for designing efficient brain-machine interfaces.
• To develop a method for mapping neuronal activity to hand position using local linear models.
• To compare the proposed strategy with existing linear and nonlinear modeling approaches.

Main Methods:

• A nonlinear combination of competitively trained local linear models (experts) was employed.
• The mapping from neuronal activity to hand position was identified.
• Training utilized normalized LMS for local models and a smaller nonlinear network for expert combination.
• Performance was compared against time-delay neural networks and recursive multilayer perceptrons.

Main Results:

• The proposed strategy achieved performance comparable to large, fully nonlinear networks.
• Significant savings in computational requirements were observed.
• Efficient combination of local linear model predictions was demonstrated.

Conclusions:

• The divide-and-conquer strategy offers an efficient alternative for BMI design.
• This approach balances computational cost and performance effectively.
• The method holds promise for developing more accessible and practical BMIs.