You are training a model on a dataset containing confidential information. To protect individual privacy and minimize the influence of any single user's data on the aggregated results, you must apply a differential privacy parameter. What should you do?