Real-time Algorithms for Sparse Neuronal System Identification.

Area of Science:

• Computational Neuroscience
• Signal Processing
• Machine Learning

Background:

• Accurate modeling of neuronal systems is crucial for understanding brain function.
• Estimating time-varying parameters from neural spiking data presents significant challenges.
• Existing methods often struggle with sparsity and real-time adaptation.

Purpose of the Study:

• To develop novel algorithms for sparse adaptive neuronal system identification.
• To enable online estimation of time-varying neuronal model parameters from spiking observations.
• To enhance the accuracy and trackability of neuronal models compared to current approaches.

Main Methods:

• Development of two adaptive filters utilizing greedy estimation techniques.
• Application of regularized log-likelihood maximization for parameter estimation.
• Validation using both simulated neural spiking data and experimental recordings from ferret auditory cortex.

Main Results:

• The proposed algorithms demonstrate significant performance gains in sparse identification.
• Improved trackability of time-varying neuronal parameters was observed.
• Outperformed existing algorithms in both simulated and real-world experimental data.

Conclusions:

• The developed adaptive filters offer a robust solution for sparse neuronal system identification.
• These methods provide efficient online estimation of dynamic neuronal parameters.
• The findings have implications for real-time analysis of neural activity and brain-computer interfaces.