Tag
This paper proposes an analytical framework for estimating the optimal early stopping time in neural network training using Rademacher complexity with L1-norm, avoiding probabilistic assumptions and applicable to both linear and nonlinear models.