CompTIA SecAI+ (CY0-001) Practice Exam

Memorization of training data in models poses privacy risks because the model can encode specific records, potentially revealing sensitive information. The most effective mitigation combines differential privacy, regularization, and leakage monitoring. Differential privacy introduces noise into the training process so that the presence or absence of a single data point has a limited effect on the model’s parameters and outputs, making it difficult to infer exact records. Regularization helps prevent the model from overfitting to idiosyncratic details, steering it toward general patterns that apply to new data. Monitoring for leakage evaluates the model for vulnerabilities like membership inference or unintended data reconstructions and flags issues so you can adjust privacy settings or retrain as needed. Other options don’t address privacy as robustly: increasing model size can heighten memorization risk, feature scaling doesn’t reduce memorization, and data augmentation alone improves generalization but doesn’t guarantee privacy protection.