Comparing Methods for Estimating Marginal Likelihood in Symbolic Regression
Explore the source record for details and available documents.
symbolic regression↗
Engineering topics
Gabriel Kronberger
Publications and source records attributed to Gabriel Kronberger.
Bayesian Symbolic Regression: Addressing Challenges in Estimating Fractional Bayes Factors and Application to Fatigue Crack Growth Modeling
This research pioneers advancements in computational mechanics by integrating Bayesian-based uncertainty quantification into symbolic regression, specifically focusing on the critical task of accurately estimating the fractional Bayes factor for selecting arbitrary equations. In our exploration, we rigorously study two prominent methods—sequential Monte Carlo and the Laplace approximation—employed for computing the fractional Bayes factor. Our findings underscore the limitations of the Laplace approximation, revealing its diminished accuracy in nonlinear and multimodal scenarios. Specifically, the Laplace approximation is shown to underpredict fractional Bayes factor on a wide set of equations associated with a symbolic regression benchmark. This comparative analysis sheds light on the nuanced performance of these techniques, guiding researchers toward more informed choices in uncertainty quantification within symbolic regression. Furthermore, we showcase the practical utility of these enhanced symbolic regression tools through their application to a real-world problem in fatigue crack growth modeling, emphasizing their efficacy in capturing the complexities of mechanical systems.
Geoffrey Bomarito↗