Privacy-Preserving Experiment Design Questions
Techniques and considerations for designing experiments and data collection strategies that protect privacy. Covers methods such as differential privacy, secure aggregation, federated learning, synthetic data, data minimization, consent management, de-identification, and privacy risk assessment, with emphasis on maintaining data utility and regulatory compliance while enabling robust experimentation.
MediumSystem Design
61 practiced
Design a privacy-preserving A/B testing service for a consumer product with 1M monthly active users. Requirements: support 500 concurrent experiments, enforce per-user epsilon budgets, provide reliable noisy p-values for teams, allow cross-team budget requests and reservations, and produce immutable audit trails. Describe key architecture components, how privacy enforcement is implemented, and how to scale and shard budgets across millions of users.
EasyTechnical
68 practiced
Explain membership inference attacks against machine learning models. Provide a concrete example of how an attacker can determine whether a given record was present in training data, the types of models most at risk, and describe practical defenses including DP-SGD, confidence-score clipping, regularization, and post-training calibration. How would you empirically evaluate whether a model remains vulnerable?
MediumTechnical
69 practiced
Explain how to compute confidence intervals for a mean or a proportion when the published statistic has Gaussian noise added to it for central differential privacy. Provide the formula for adjusted confidence intervals that account for both sampling variance and DP noise variance, and explain how to interpret p-values derived from noisy statistics.
MediumTechnical
65 practiced
In Python, implement a simple Rényi Differential Privacy (RDP) accountant that computes the cumulative RDP for T compositions of a Gaussian mechanism without subsampling, and then converts the RDP to an (epsilon, delta) pair for a target delta. Function signature: compute_epsilon_rdp(sigma, steps, delta). Explain numeric stability choices and validate on a small example.
EasyTechnical
74 practiced
Compare k-anonymity, l-diversity, t-closeness, pseudonymization, and differential privacy as protections for datasets used in ML experiments. For each technique describe its protection model, common attacks that can break it (for example linkage attacks), typical use cases in analytics or ML, and main limitations with respect to utility and provable guarantees.
Unlock Full Question Bank
Get access to hundreds of Privacy-Preserving Experiment Design interview questions and detailed answers.
Sign in to ContinueJoin thousands of developers preparing for their dream job.