Prototype-Rule Neurosymbolic Regularization for Rank-Constrained Tensor Neural Networks under Label Scarcity
This study investigates adding a prototype-rule regularization to Rank-R tensor neural networks to improve classification under limited supervision. The method enforces a rule that pushes an embedding toward its labeled class prototype while keeping it far from other classes. Results show this regularization helps performance, but its benefit depends heavily on the specific experimental setup and dataset.