An ML engineer uses Amazon SageMaker AI to train a deep learning model. During testing, the model exhibits high variance. The ML engineer must adjust a hyperparameter to improve generalization without significantly increasing training time. Which solution will **MOST** improve model generalization?