Product units in gated recurrent units improve nuclear-mass prediction
Researchers have developed a new machine learning technique using gated recurrent units (GRUs) to improve the prediction of atomic nuclei masses. By incorporating multiplicative interactions and product-unit transformations within the GRU architecture, the model achieved state-of-the-art results in both interpolation and extrapolation tasks. The complex-valued additive-multiplicative product-unit GRU (AM-PU-GRU) demonstrated lower prediction errors than existing machine learning models and traditional GRU baselines. AI
IMPACT Establishes a new benchmark for sequence-based nuclear mass prediction, potentially accelerating scientific discovery in nuclear physics.