Name common data privacy techniques used in machine learning.

• A. Differential privacy, data minimization, anonymization/de-identification, federated learning, and encryption.
• B. Data replication, caching, index optimization, and lazy loading.
• C. Cross-validation, hyperparameter tuning, and regularization.
• D. Data augmentation, transfer learning, fine tuning.

Answer: (A)

Privacy-preserving techniques in machine learning aim to protect individuals’ data while still enabling useful model training. The set of methods listed here covers the main ways this protection is achieved. Differential privacy adds carefully calibrated noise to data or to model outputs, so the contribution of any single person cannot be deduced, reducing leakage and re-identification risk. Data minimization focuses on collecting and retaining only what is strictly necessary, shrinking the amount of sensitive information at risk. Anonymization or de-identification removes or obfuscates direct identifiers to make it harder to link data back to individuals. Federated learning keeps data on local devices or servers and only shares model updates, which means raw data never leaves the source. Encryption protects data during storage and transmission, ensuring that unauthorized parties cannot read it even if they access the data.

These techniques address different stages of the ML lifecycle—from data collection and storage to computation and output—making them a comprehensive answer for privacy in ML. Other approaches listed are more about performance, model development, or efficiency rather than privacy protection.