Inference, Prediction, & Entropy-Rate Estimation of Continuous-Time, Discrete-Event Processes

Area of Science:

• Machine Learning
• Statistical Modeling
• Information Theory

Background:

• Established methods effectively model discrete-time processes.
• Continuous-time discrete-event processes are prevalent but less studied.
• Existing techniques for continuous-time processes have limitations.

Purpose of the Study:

• To introduce novel methods for inferring, predicting, and estimating continuous-time discrete-event processes.
• To extend Bayesian structural inference using neural network capabilities.
• To provide a more comprehensive framework for analyzing complex event data.

Main Methods:

• Extension of Bayesian structural inference.
• Integration of neural networks for universal approximation.
• Application to discrete-event processes in continuous time.

Main Results:

• The proposed methods demonstrate strong performance in inferring models.
• New capabilities for prediction in continuous-time processes are achieved.
• Entropy rate estimation for these processes is competitive with state-of-the-art.
• Experiments on complex synthetic data validate the approach.

Conclusions:

• The developed methods offer a powerful new tool for continuous-time discrete-event process analysis.
• Neural network integration enhances the capabilities of Bayesian structural inference.
• These advancements are significant for fields dealing with continuous-time event data.