Optimal Inference of Hidden Markov Models Through Expert-Acquired Data.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Systems Biology

Background:

• Traditional methods for inferring Hidden Markov models (HMMs) primarily rely on temporal data dynamics.
• Expert-acquired data, encompassing human decisions in domains like navigation, cybersecurity, and biology, offer rich insights but are often underutilized in HMM inference.
• Existing approaches lack mechanisms to effectively incorporate expert knowledge, limiting model accuracy and applicability.

Purpose of the Study:

• To develop a novel method for HMM inference that integrates expert knowledge alongside temporal data.
• To model expert behavior as an imperfect reinforcement learning agent to quantify their understanding of the system.
• To demonstrate the method's effectiveness across various inference criteria and complex real-world applications.

Main Methods:

• Incorporation of expert knowledge by modeling their decision-making process using reinforcement learning principles.
• Development of an inference method combining dynamic programming and optimal recursive Bayesian estimation.
• Quantification of expert perceptions of the system model to augment temporal data analysis.

Main Results:

• The proposed method effectively integrates expert insights into HMM inference, outperforming conventional approaches.
• Demonstrated applicability to diverse inference criteria, including maximum likelihood and maximum a posteriori.
• Validation through comprehensive numerical experiments on a benchmark problem and biological networks, showcasing robust performance.

Conclusions:

• Expert knowledge can be effectively leveraged to enhance HMM inference, leading to more accurate system models.
• The proposed method provides a principled way to combine temporal data with imperfect expert guidance.
• This approach holds significant potential for applications in autonomous systems, cybersecurity, and biological network analysis.