Kolmogorov-Arnold networks in nuclear binding energy prediction
Kolmogorov-Arnold networks (KANs) were tested to predict nuclear binding energies using features like atomic and neutron numbers. By incorporating additional physical properties, KANs achieved a low root mean square error of 0.26 MeV on the entire dataset, outperforming simpler models. Symbolic regression also yielded an analytical expression for binding energy, suggesting improved theoretical understanding.