Modeling a recurrent, hidden dynamical system using energy minimization and kernel density estimates.

Area of Science:

• Dynamical Systems
• Statistical Modeling
• Machine Learning

Background:

• Trajectory data is often sparse and noisy, posing challenges for accurate modeling.
• Kernel Density Estimation (KDE) has shown promise in analyzing complex datasets, including trajectory data.
• Previous work has utilized KDE for anomaly detection in shipping data and trajectory modeling.

Purpose of the Study:

• To develop a novel Kernel Density Estimation (KDE) approach for modeling and forecasting recurrent trajectories.
• To address the challenges of sparse sampling and observational noise in trajectory reconstruction.
• To demonstrate the effectiveness of the proposed KDE method in handling imperfect trajectory data.

Main Methods:

• Development of a Kernel Density Estimation (KDE) framework tailored for manifold-based trajectory analysis.
• Application of KDE to model sequences of coordinates in a phase space governed by hidden dynamical systems.
• Focus on reconstructing trajectories from sparsely sampled and noisy data.

Main Results:

• The proposed KDE approach provides a robust method for modeling and forecasting recurrent trajectories.
• The technique successfully reconstructs trajectories even with sparse and noisy data.
• Theoretical analysis shows the constructed estimator minimizes a specific energy function as sample size increases.

Conclusions:

• Kernel Density Estimation offers a powerful tool for analyzing and predicting complex trajectories.
• The developed method advances the state-of-the-art in sparse and noisy trajectory reconstruction.
• This work has implications for understanding and forecasting systems with hidden dynamics.