Discovering Nuclear Models from Machine Learning.
In this work, we present how Machine Learning models, such as Fully Connected Neural Networks (FC Nets) and Symbolic Regressions, can aid in deriving insights from nuclear data. We have demonstrated that FC Nets are effective in predicting nuclear data trends across the entire nuclear chart with minimal error. However, these models are insufficient in capturing the underlying physics at the level of local isotopic chains. To address this, we explored the use of Symbolic Regression as a tool to overcome such limitations. Despite its potential, our current model requires further modifications to align its predictions with established phenomenological models, such as the Liquid Drop Model of nuclei. Recent advancements in Symbolic Regression techniques for nuclear physics data shows hope in enhancing both theoretical and experimental understanding. These developments can help address key questions from new perspectives, paving the way for improved modeling and analysis.
Acknowledgement https://github.com/hbprosper/MLinPhysics.git