A hierarchical Bayesian inference model for volatile multivariate exponentially distributed signals

Area of Science:

• Computational Neuroscience
• Statistical Modeling
• Information Theory

Background:

• Brain activity often exhibits exponential distributions, posing challenges for data analysis due to memoryless and peakless properties.
• Estimating rate parameters in multivariate exponential distributions from time-series sensory data is complex.
• Existing methods struggle with the intricate interactions within multivariate exponential random variables.

Purpose of the Study:

• To develop a robust method for estimating the rate parameter of multivariate exponential distributions from time-series sensory inputs.
• To address the difficulties imposed by the memoryless and peakless properties of exponential distributions in data analysis.
• To create a model capable of analyzing high-dimensional neural activities by accounting for complex interactions.

Main Methods:

• Construction of a hierarchical Bayesian inference model utilizing a variant of the general hierarchical Brownian filter (GHBF).
• Estimation of the second-order interaction of the rate intensity parameter in logarithmic space to handle complex interactions.
• Application of a variational Bayesian scheme to derive closed-form and analytical update equations.

Main Results:

• The developed model successfully evaluates time-varying rate parameters of multivariate exponential distributions.
• The model accurately identifies the underlying correlation structure of volatile multivariate exponentially distributed signals.
• Simulation studies validate the model's capability in analyzing complex neural data.

Conclusions:

• The proposed hierarchical Bayesian inference model offers a practical solution for analyzing high-dimensional neural activities.
• The model's predictive coding framework and analytical update equations enhance the analysis of exponentially distributed signals.
• This approach provides a powerful tool for understanding the dynamics of neural processes.