Covariance density estimation for autoregressive spectral modelling of point processes

Area of Science:

• Time Series Analysis
• Point Process Statistics
• Spectral Analysis

Background:

• Autoregressive (AR) modeling is vital for time series analysis and potentially for point processes.
• Existing AR spectral analysis methods require positive semidefinite covariance function estimates.
• Current point process covariance density estimation methods do not guarantee positive semidefinite estimates.

Purpose of the Study:

• To develop and present computational efficient algorithms for estimating point process covariance density and conditional intensity functions.
• To ensure these estimation algorithms yield positive semidefinite estimates suitable for AR spectral analysis.
• To apply these methods to neurobiological data for AR spectral modeling.

Main Methods:

• Discusses methods for estimating covariance density and conditional intensity functions of point processes.
• Presents alternative computational efficient estimation algorithms.
• Utilizes Yule-Walker type equations and Levinson recursion for AR spectral modeling.
• Employs the minimum Akaike Information Criterion (AIC) or Consistent Akaike Information Criterion (CAT) principle.

Main Results:

• Developed estimation algorithms that consistently produce positive semidefinite covariance density estimates.
• Demonstrated the adequacy of these estimates for autoregressive spectral analysis of point processes.
• Successfully applied the AR spectral modeling technique to neurobiological data.

Conclusions:

• The proposed methods provide a robust approach to autoregressive spectral analysis for point processes.
• Ensuring positive semidefinite estimates overcomes a key limitation in applying AR modeling to point process data.
• The study highlights the utility of these techniques in analyzing complex neurobiological data.