Assess learners' ability to apply differential privacy principles to machine learning training: understanding ε/δ and sensitivity, selecting and configuring mechanisms (Gaussian/Laplace noise, randomized response), implementing algorithms (DP‑SGD, objective/output perturbation), using privacy accounting methods (moments accountant, RDP) to compute cumulative privacy loss, choosing clipping and noise scale to meet target (ε,δ) while managing utility-privacy tradeoffs, and evaluating impacts on model performance, fairness, and deployment. Tasks include designing a DP training plan for a specified dataset, calculating required noise and expected utility change, and justifying implementation choices with practical constraints.