Sharpness-Aware Minimization with Adaptive Regularization for Training Deep Neural Networks