Sparse Ensemble Machine Learning to Improve Robustness of Long-Term Decoding in iBMIs.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Intracortical Brain-Machine Interfaces (iBMIs) face challenges due to non-stationary neural data distributions over time.
• Robustness is crucial for reliable iBMI performance in real-world applications.

Purpose of the Study:

• To develop and evaluate a novel sparse ensemble machine learning approach to enhance the robustness of iBMIs.
• To assess the performance improvements across various base classifiers.

Main Methods:

• A sparse ensemble machine learning technique was proposed, where each classifier is trained on a randomly sampled subset of input channels.
• The approach was tested using Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and Multilayer Perceptron (MLP) as base classifiers.
• The method was validated on non-human primate (NHP) datasets.

Main Results:

• The sparse ensemble approach demonstrated significant improvements in decoding accuracy compared to single classifier models across different algorithms.
• Accuracy gains of up to ≈21% in NHP A and 7% in NHP B were observed for LDA.
• Improvements over Random Forest (Long-short Term Memory) were also noted, reaching up to ≈15% in NHP B for sparse ensemble MLP.

Conclusions:

• The proposed sparse ensemble method effectively enhances the robustness and decoding accuracy of iBMIs in the presence of non-stationary neural data.
• This approach offers a promising strategy for improving the reliability of brain-computer interfaces.