Researchers B. Maheshwari and P. Van Isacker, affiliated with the University of Lyon in France, have developed a novel machine learning (ML) framework that aims to uncover underlying physical principles from complex datasets. Their work, published in the journal Physical Review Letters, focuses on understanding nuclear charge radii, but the methodology could have broader implications for the energy sector, particularly in nuclear energy research and development.
The researchers introduced a hybrid ML approach that combines numerical regression for high-accuracy predictions with symbolic regression to derive analytical expressions. This method leverages the smoothed predictions of optimized ML models, significantly accelerating the process compared to direct symbolic regression on raw experimental data. By using Light Gradient Boosting Machine (LGBM) and Gaussian Process Regression (GPR) models, they mapped correlations between charge radii and key physical features, including mass and proton number dependencies, total binding energy, and, for the first time, the pairing gap.
The models were rigorously trained using four-fold cross-validation with automated hyperparameter optimization, ensuring robustness and generalizability. This is particularly important given the typically small and skewed datasets in nuclear physics. The researchers then distilled the knowledge from the initial LGBM and GPR models into simplified, interpretable mathematical expressions via symbolic regression, effectively “white-boxing” these ML models. The derived formulas provide physical insights comparable to traditional many-body models and demonstrate a powerful pathway for physics expression discovery guided by ML.
For the energy sector, this research highlights the potential of ML to uncover new physical insights and optimize processes in nuclear energy. By understanding nuclear charge radii and other key physical features, researchers can improve nuclear reactor designs, enhance nuclear fuel cycles, and develop more accurate models for nuclear waste management. The methodology could also be applied to other areas of energy research, such as understanding material properties for advanced energy storage systems or optimizing renewable energy technologies.
In summary, Maheshwari and Van Isacker’s work showcases the power of ML in uncovering underlying physical principles from complex datasets. Their hybrid approach not only provides high-accuracy predictions but also derives interpretable mathematical expressions, offering valuable insights for the energy sector and beyond. The research was published in Physical Review Letters, a prestigious journal in the field of physics.
This article is based on research available at arXiv.