Fast Kalman filtering on quasilinear dendritic trees

Area of Science:

• Computational cellular neuroscience
• Biophysics
• Signal processing

Background:

• Accurate filtering of noisy voltage signals in dendritic trees is crucial for understanding neuronal function.
• High-dimensional state spaces in multicompartmental models (N ≈ 10^4) make standard Kalman filters computationally intractable (O(N^3) time, O(N^2) space).

Purpose of the Study:

• To develop an efficient filtering method for high-dimensional dendritic voltage signals.
• To overcome the computational limitations of standard Kalman filters for large-scale neuronal models.

Main Methods:

• Leveraged tree structure of dendritic dynamics and sparse matrix methods for O(N) cable equation solutions.
• Approximated Kalman equations using a low-rank perturbation of the steady-state solution.
• Exploited sparse tree structure for O(N) steady-state solution computation.

Main Results:

• Developed an efficient filter requiring only O(N) time and space, a significant improvement over standard Kalman filters.
• The proposed method provides a highly accurate approximation to the exact Kalman filter solution.
• Demonstrated applicability to real and simulated dendritic structures, with extensions for various observation types.

Conclusions:

• The efficient filter enables practical analysis of voltage dynamics in complex neuronal models.
• This approach overcomes computational bottlenecks, facilitating advanced research in computational neuroscience.
• The method is adaptable to diverse experimental data and observation modalities.