Online Estimating Pairwise Neuronal Functional Connectivity in Brain-Machine Interface.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Neurons form networks through pairwise interactions, crucial for encoding movement.
• Brain-machine interfaces (BMIs) investigate neural adaptation and connectivity.
• Estimating high-dimensional, non-stationary neural states from spike trains is challenging.

Purpose of the Study:

• To develop an online method for simultaneously estimating functional neural connectivity and neuronal tuning adaptation.
• To address the challenges of high dimensionality and non-stationarity in neural state estimation.

Main Methods:

• An Adam-based gradient descent algorithm was employed for online estimation.
• The method minimizes negative log-likelihood using point process observations.
• Adaptive learning rates, momentum, and regularization were utilized for neural state vectors.

Main Results:

• The proposed method accurately estimates high-dimensional pairwise neuronal functional connectivity and single neuronal tuning adaptation.
• It outperforms existing methods, particularly in sparse state conditions.
• The method demonstrates superior stability and speed in online scenarios, irrespective of parameter initialization.

Conclusions:

• The developed method offers a robust tool for tracking time-variant functional neural connectivity.
• It facilitates the dynamic formation of functional neural networks for enhanced brain control.
• This approach promises to advance the capabilities of brain-machine interfaces.