You've noticed that a machine learning model's performance in production doesn't match its validation test results, leading you to suspect overfitting. Which training technique can help mitigate this risk?