Differential privacy techniques provide statistical guarantees to prevent trained machine learning models from memorizing and exposing sensitive individual data