×

Trending Technologies MCQs

Differential Privacy MCQs (Multiple-Choice Questions)

Differential Privacy is a mathematical framework for protecting individual privacy when analyzing or sharing information derived from datasets. It introduces carefully calibrated randomness into computations or statistical outputs to limit how much the results reveal about any one individual's participation. Differential privacy is widely used in statistical analysis, machine learning, public data reporting, and privacy-preserving research.

Differential Privacy MCQs

These Differential Privacy multiple-choice questions cover the fundamental principles, mathematical definitions, privacy parameters, mechanisms, and practical applications of differential privacy. They are suitable for students, data scientists, cybersecurity professionals, machine learning engineers, and researchers interested in statistical privacy.

Each question includes four answer options, the correct answer, and an explanation. The questions progress from basic concepts to advanced topics such as epsilon and delta, sensitivity, neighboring datasets, the Laplace mechanism, the Gaussian mechanism, privacy budgets, composition, randomized response, and privacy-preserving machine learning.

List of Differential Privacy MCQs

The following MCQs cover differential privacy definitions, privacy guarantees, query sensitivity, noise calibration, Laplace and Gaussian mechanisms, local and central differential privacy, randomized response, privacy loss, composition, privacy accounting, clipping, data aggregation, and practical calculation scenarios.

1. What is the primary objective of differential privacy?

  1. To eliminate all errors from a dataset
  2. To make every record publicly identifiable
  3. To limit how much an analysis reveals about an individual's participation or data
  4. To encrypt all data permanently so it cannot be analyzed

Answer: C) To limit how much an analysis reveals about an individual's participation or data

Explanation:

Differential privacy provides a mathematical framework for limiting the influence of an individual record on the output of an analysis. It helps reduce the risk that someone can infer sensitive information about a person from the released results.

2. What is the central idea behind differential privacy?

  1. Results should remain similar whether one individual's data is included or excluded, within a defined privacy guarantee
  2. Every query must return exactly the same result
  3. All dataset records must be encrypted with the same password
  4. Data analysts must never calculate statistical averages

Answer: A) Results should remain similar whether one individual's data is included or excluded, within a defined privacy guarantee

Explanation:

Differential privacy bounds how much the output distribution can change between appropriately defined neighboring datasets. This makes it harder to determine whether a particular individual contributed to the data based solely on the released result.

3. What does the privacy parameter epsilon (ε) represent?

  1. The total number of records in a database
  2. The amount of storage required by an algorithm
  3. The number of queries executed per second
  4. A parameter that controls the strength of the differential privacy guarantee

Answer: D) A parameter that controls the strength of the differential privacy guarantee

Explanation:

Epsilon controls how much the output distributions may differ between neighboring datasets. For the same privacy definition and assumptions, a smaller epsilon generally provides a stronger privacy guarantee, often at the cost of reduced statistical accuracy.

4. What does the delta (δ) parameter represent in approximate differential privacy?

  1. The number of columns in a dataset
  2. An additional probability allowance in the privacy guarantee
  3. The exact number of records that must be removed
  4. The accuracy percentage of every released statistic

Answer: B) An additional probability allowance in the privacy guarantee

Explanation:

In (ε, δ)-differential privacy, delta provides an additive allowance in the probability inequality defining the guarantee. When choosing delta, practitioners should consider the dataset size, threat model, and applicable privacy requirements.

5. What are neighboring datasets in differential privacy?

  1. Datasets that use the same file format
  2. Datasets stored on adjacent servers
  3. Datasets that differ by the presence or absence of one individual's data, or by another explicitly defined neighboring relation
  4. Datasets that always contain identical values

Answer: C) Datasets that differ by the presence or absence of one individual's data, or by another explicitly defined neighboring relation

Explanation:

Neighboring datasets are defined to differ in a limited way, commonly through the addition or removal of one individual's record. Some definitions instead use replacement of one record. The chosen relation affects the interpretation and calibration of the privacy guarantee.

6. Which mechanism is commonly used to achieve differential privacy for numerical queries?

  1. Laplace mechanism
  2. Plain-text encoding
  3. File compression
  4. Database indexing

Answer: A) Laplace mechanism

Explanation:

The Laplace mechanism adds random noise drawn from a Laplace distribution to a query result. The noise scale is calibrated using the query's sensitivity and the selected epsilon parameter.

7. What is query sensitivity in differential privacy?

  1. The number of users who access a query
  2. The time required to execute a query
  3. The amount of memory needed to store query results
  4. The maximum change in a query's output between neighboring datasets, under a specified measure

Answer: D) The maximum change in a query's output between neighboring datasets, under a specified measure

Explanation:

Sensitivity measures how much a query result can change when the dataset changes according to the chosen neighboring relation. It is a key input when determining how much random noise must be added to achieve a desired privacy guarantee.

8. What is the primary purpose of adding random noise in differential privacy?

  1. To make every dataset larger
  2. To reduce the ability to infer information about any one individual's contribution
  3. To ensure every result is exactly correct
  4. To remove the need for statistical analysis

Answer: B) To reduce the ability to infer information about any one individual's contribution

Explanation:

Random noise makes outputs less dependent on any single individual's data. The amount and distribution of noise are chosen according to the query sensitivity, privacy parameters, and mechanism being used.

9. Which statement about smaller epsilon values is generally correct?

  1. They generally provide stronger privacy guarantees under the same definition, but may require more noise
  2. They always increase the exact accuracy of released statistics
  3. They eliminate all computational overhead
  4. They guarantee that no one can infer any information from the output

Answer: A) They generally provide stronger privacy guarantees under the same definition, but may require more noise

Explanation:

A smaller epsilon generally limits how much output distributions may differ between neighboring datasets. Achieving this stronger guarantee often requires adding more noise, which can reduce the utility of the released results.

10. What is the Laplace distribution used for in the Laplace mechanism?

  1. To encrypt the complete database
  2. To determine the number of database tables
  3. To generate noise with a scale calibrated to query sensitivity and privacy requirements
  4. To replace every record with a random identifier

Answer: C) To generate noise with a scale calibrated to query sensitivity and privacy requirements

Explanation:

The Laplace mechanism adds noise drawn from a Laplace distribution. For a scalar query with L1 sensitivity Δf, a common calibration uses scale b = Δf/ε to achieve ε-differential privacy under the standard assumptions.

11. Which mechanism is commonly associated with Gaussian noise?

  1. Base64 mechanism
  2. Gaussian mechanism
  3. Sorting mechanism
  4. Compression mechanism

Answer: B) Gaussian mechanism

Explanation:

The Gaussian mechanism adds noise sampled from a normal distribution. It is commonly used to achieve approximate differential privacy, with noise calibrated according to the query's L2 sensitivity and the selected privacy parameters.

12. What is local differential privacy?

  1. A method that requires all user records to be stored on one server
  2. A privacy framework that only applies to geographic data
  3. A technique that guarantees exact answers without noise
  4. A model in which data is randomized on the user's device or before being shared with the data collector

Answer: D) A model in which data is randomized on the user's device or before being shared with the data collector

Explanation:

Local differential privacy allows a client to randomize information before sending it to a collector. The collector does not need direct access to the original value, although the approach can require more noise and larger sample sizes for useful analysis.

13. What is central differential privacy?

  1. A model in which a trusted curator holds the dataset and releases differentially private outputs
  2. A method that prohibits all data collection
  3. A system that stores every record on a user's device only
  4. A technique that removes all randomization from query results

Answer: A) A model in which a trusted curator holds the dataset and releases differentially private outputs

Explanation:

In the central model, a curator processes the dataset and applies a differentially private mechanism before releasing results. This model can provide useful statistical accuracy, but it relies on protecting the raw dataset and trusting the curator's handling of it.

14. What is randomized response?

  1. A method for permanently deleting database records
  2. A technique that always reports the true answer
  3. A survey technique that randomizes responses to provide plausible deniability and can support local differential privacy
  4. A method for encrypting files with a public key

Answer: C) A survey technique that randomizes responses to provide plausible deniability and can support local differential privacy

Explanation:

Randomized response introduces controlled randomness into answers before they are collected. With an appropriately designed mechanism, analysts can estimate population statistics while reducing the ability to determine an individual's true response.

15. What is a privacy budget in differential privacy?

  1. The monetary cost of buying a database
  2. A limit on the cumulative privacy loss permitted across analyses or queries
  3. The amount of storage allocated to encrypted data
  4. The number of passwords a user can create

Answer: B) A limit on the cumulative privacy loss permitted across analyses or queries

Explanation:

A privacy budget tracks how privacy guarantees are consumed across releases. Repeated analyses can accumulate privacy loss, so systems often use privacy accounting to determine whether additional queries fit within the permitted budget.

16. What does composition mean in differential privacy?

  1. Combining multiple files into one archive
  2. Encrypting several passwords with the same key
  3. Removing duplicate rows from a database
  4. Combining privacy guarantees from multiple mechanisms or releases to determine the overall privacy loss

Answer: D) Combining privacy guarantees from multiple mechanisms or releases to determine the overall privacy loss

Explanation:

Composition describes how privacy loss accumulates when multiple differentially private analyses are performed. Basic composition provides a straightforward bound, while advanced composition and other accounting methods can provide tighter bounds under appropriate assumptions.

17. Which statement about differential privacy and data encryption is correct?

  1. Differential privacy limits disclosure through outputs, while encryption protects data by restricting access to its contents
  2. They are exactly the same cryptographic technique
  3. Encryption automatically provides differential privacy
  4. Differential privacy always requires public encryption keys

Answer: A) Differential privacy limits disclosure through outputs, while encryption protects data by restricting access to its contents

Explanation:

Encryption protects information from parties that do not possess the appropriate decryption capability. Differential privacy limits the influence of individual records on released outputs. The techniques address different risks and can be used together.

18. What is the main purpose of clipping in differentially private machine learning?

  1. To increase every model parameter without a limit
  2. To remove the need for a training dataset
  3. To bound the influence of individual examples or their gradients
  4. To guarantee that a model never makes incorrect predictions

Answer: C) To bound the influence of individual examples or their gradients

Explanation:

Clipping limits the magnitude of contributions, such as per-example gradients, so that sensitivity can be bounded. Differentially private training then adds appropriately calibrated noise and accounts for privacy loss over training steps.

19. What is differential privacy's post-processing property?

  1. Every processed result must be more accurate than the original result
  2. Processing a differentially private output using a function that does not access the private dataset does not worsen its privacy guarantee
  3. All output values must be rounded to zero
  4. Every additional computation consumes an equal amount of privacy budget

Answer: B) Processing a differentially private output using a function that does not access the private dataset does not worsen its privacy guarantee

Explanation:

The post-processing property states that transformations of a differentially private output do not weaken its established privacy guarantee, provided they do not access the original private data or introduce additional unaccounted data-dependent releases.

20. Which of the following is a practical application of differential privacy?

  1. Publishing every user's password
  2. Removing all randomness from statistical reports
  3. Allowing unrestricted access to confidential records
  4. Releasing population statistics while limiting the influence of individual records

Answer: D) Releasing population statistics while limiting the influence of individual records

Explanation:

Differential privacy can be used to publish statistics, such as counts or averages, while limiting how much an individual's data affects the output. It is useful when organizations need aggregate insights without exposing individual contributions.

21. What is the relationship between sensitivity and the amount of noise required?

  1. Higher sensitivity generally requires more noise for the same privacy parameters
  2. Higher sensitivity always requires less noise
  3. Sensitivity has no relationship to noise calibration
  4. Noise depends only on the database file name

Answer: A) Higher sensitivity generally requires more noise for the same privacy parameters

Explanation:

When a query can change substantially between neighboring datasets, stronger noise is generally required to achieve the same privacy guarantee. Bounding sensitivity is therefore an important step in designing useful differentially private mechanisms.

22. Which statement about the accuracy of differentially private results is correct?

  1. Differential privacy guarantees exact answers for all queries
  2. Adding noise always makes the result more accurate
  3. Noise can reduce accuracy, creating a trade-off between privacy and statistical utility
  4. Accuracy is unrelated to the choice of privacy parameters

Answer: C) Noise can reduce accuracy, creating a trade-off between privacy and statistical utility

Explanation:

Differential privacy introduces controlled randomness that can affect the accuracy of released statistics. Stronger privacy settings often require more noise, so practitioners must balance privacy requirements against the utility needed for a particular task.

23. What is the main purpose of privacy accounting in differentially private machine learning?

  1. To count the number of model layers
  2. To measure how privacy loss accumulates during training and other data-dependent operations
  3. To calculate the physical size of a server
  4. To guarantee perfect model accuracy

Answer: B) To measure how privacy loss accumulates during training and other data-dependent operations

Explanation:

Privacy accounting tracks the cumulative privacy cost of operations such as repeated training steps. Methods such as Rényi differential privacy accounting can provide useful bounds for analyzing the overall privacy guarantee.

24. What is Rényi differential privacy (RDP) commonly used for?

  1. Encrypting database backups
  2. Replacing all statistical analysis
  3. Converting every data record into a password
  4. Analyzing and accounting for privacy loss, particularly when composing multiple mechanisms

Answer: D) Analyzing and accounting for privacy loss, particularly when composing multiple mechanisms

Explanation:

Rényi differential privacy provides a framework for describing privacy guarantees that can simplify composition analysis. It is commonly used in privacy accounting for iterative algorithms, including differentially private machine learning.

25. What does pure epsilon-differential privacy mean?

  1. The privacy guarantee uses epsilon without an additional delta allowance
  2. The dataset contains only numerical values
  3. The algorithm never introduces randomness
  4. The output must always equal the exact query result

Answer: A) The privacy guarantee uses epsilon without an additional delta allowance

Explanation:

Pure ε-differential privacy bounds the output probabilities for neighboring datasets using a multiplicative factor of \(e^\varepsilon\), without an additive delta term. It is a specific form of differential privacy with its own privacy and utility trade-offs.

26. Why can releasing the same differentially private result repeatedly still be a privacy concern?

  1. Every repeated output automatically reveals all raw records
  2. Repeated outputs always cancel each other's noise
  3. Repeated independent releases can accumulate privacy loss
  4. Repeated outputs eliminate the need for a privacy budget

Answer: C) Repeated independent releases can accumulate privacy loss

Explanation:

Repeated releases of related statistics can increase cumulative privacy loss, even when each release is individually differentially private. A privacy accountant should track the combined guarantee, and simply repeating a query is not a substitute for managing the privacy budget.

27. What is one advantage of local differential privacy?

  1. It allows a collector to recover every user's original value exactly
  2. It can protect raw user values before they are transmitted to the data collector
  3. It eliminates the need to choose a privacy parameter
  4. It always provides more accurate statistics than central differential privacy

Answer: B) It can protect raw user values before they are transmitted to the data collector

Explanation:

Local differential privacy randomizes data on the user's device or before it reaches the collector. This reduces the collector's access to exact individual responses, but often requires larger datasets to achieve useful statistical accuracy.

28. What is a common challenge when applying differential privacy to small datasets?

  1. It is impossible to calculate any statistic
  2. The dataset cannot contain numerical values
  3. Differential privacy always reveals every individual's identity
  4. The required noise may be large relative to the dataset, reducing the usefulness of results

Answer: D) The required noise may be large relative to the dataset, reducing the usefulness of results

Explanation:

When datasets are small, the noise required for a given privacy guarantee may substantially affect the released statistic. Careful query design, sensitivity reduction, and appropriate use-case selection can help improve utility without weakening the intended privacy guarantee.

29. Which technique can help make the sensitivity of a numerical query manageable?

  1. Bounding each individual's contribution to the query
  2. Allowing one individual to contribute an unlimited number of records
  3. Publishing the raw dataset before analysis
  4. Disabling privacy accounting

Answer: A) Bounding each individual's contribution to the query

Explanation:

Bounding contributions prevents one individual from having an arbitrarily large influence on the result. Examples include limiting the number of records per person or clipping numerical values to a specified range before computing a statistic.

30. What is one limitation of differential privacy?

  1. It cannot be applied to numerical data
  2. It always requires data to be publicly available
  3. It does not automatically prevent every form of data misuse or protect information exposed outside the analyzed outputs
  4. It eliminates the need for secure system design

Answer: C) It does not automatically prevent every form of data misuse or protect information exposed outside the analyzed outputs

Explanation:

Differential privacy provides a formal guarantee for specified data-dependent outputs. It does not replace access controls, encryption, secure data storage, appropriate governance, or protection against unrelated information leaks.

31. What does group privacy describe in differential privacy?

  1. A guarantee that all members of a group have identical records
  2. How the privacy guarantee changes when neighboring datasets differ in multiple individuals' records
  3. A rule requiring every dataset to have the same number of rows
  4. A technique for storing data in group folders

Answer: B) How the privacy guarantee changes when neighboring datasets differ in multiple individuals' records

Explanation:

Group privacy extends individual-level guarantees to changes involving multiple records. Under basic pure differential privacy composition, the epsilon bound can scale with the number of changed individuals, illustrating why protecting a group may require a different interpretation from protecting one person.

32. What is the main role of the Gaussian mechanism's standard deviation?

  1. To determine the number of columns in a dataset
  2. To remove the need for a privacy parameter
  3. To calculate a database's physical size
  4. To control the magnitude of added Gaussian noise, calibrated to sensitivity and privacy requirements

Answer: D) To control the magnitude of added Gaussian noise, calibrated to sensitivity and privacy requirements

Explanation:

The Gaussian mechanism adds normally distributed noise whose standard deviation is selected according to the query's L2 sensitivity and the desired privacy guarantee. Exact calibration depends on the differential privacy definition and mechanism being used.

33. Why is data contribution bounding important in differentially private analytics?

  1. It limits how much one individual can affect the query result
  2. It guarantees that all query outputs are exact
  3. It eliminates the need for random noise
  4. It prevents analysts from using numerical data

Answer: A) It limits how much one individual can affect the query result

Explanation:

Contribution bounding restricts how much an individual can influence a query. It helps establish a manageable sensitivity bound, which is necessary for calibrating noise and achieving a well-defined differential privacy guarantee.

34. Which statement about differential privacy and machine learning models is correct?

  1. Differential privacy guarantees that every prediction is correct
  2. Differential privacy makes model evaluation unnecessary
  3. Differentially private training can limit information leakage about training examples, but may affect model utility
  4. Differential privacy requires publishing the training dataset

Answer: C) Differentially private training can limit information leakage about training examples, but may affect model utility

Explanation:

Differentially private training techniques, such as DP-SGD, can bound the influence of individual training examples and add noise to model updates. These methods provide a privacy guarantee under specified assumptions, but the noise and privacy parameters may affect predictive performance.

35. What is DP-SGD?

  1. A database encryption standard
  2. Differentially Private Stochastic Gradient Descent, which typically clips per-example gradients and adds calibrated noise
  3. A method for sorting encrypted records
  4. A data compression algorithm for image files

Answer: B) Differentially Private Stochastic Gradient Descent, which typically clips per-example gradients and adds calibrated noise

Explanation:

DP-SGD adapts stochastic gradient descent to support differential privacy. It typically clips per-example gradients, adds appropriately calibrated random noise to aggregated gradients, and tracks cumulative privacy loss across training iterations.

36. What is the purpose of privacy amplification by subsampling?

  1. To make every training record publicly visible
  2. To eliminate the need for noise in every mechanism
  3. To guarantee that every individual is selected in each iteration
  4. To obtain a stronger privacy bound in certain settings by randomly selecting only a subset of data for a mechanism

Answer: D) To obtain a stronger privacy bound in certain settings by randomly selecting only a subset of data for a mechanism

Explanation:

Subsampling can amplify privacy under appropriate assumptions because an individual may not be included in a particular sample. The resulting guarantee depends on the sampling method, the underlying mechanism, and the privacy accountant used.

37. Which statement about differential privacy and anonymization is correct?

  1. Differential privacy provides a formal bound on information revealed by specified outputs, while removing names alone does not guarantee anonymity
  2. Removing names from a dataset always guarantees differential privacy
  3. Differential privacy requires every record to be deleted
  4. Anonymization and differential privacy are identical in every implementation

Answer: A) Differential privacy provides a formal bound on information revealed by specified outputs, while removing names alone does not guarantee anonymity

Explanation:

Removing direct identifiers may leave information that can be linked to other datasets. Differential privacy instead provides a mathematical guarantee about output distributions for neighboring datasets, although it does not eliminate every possible privacy risk.

38. What is one reason to use a differentially private count query?

  1. To expose every individual record in the dataset
  2. To avoid all statistical uncertainty
  3. To release an approximate population count while limiting the influence of individual records
  4. To guarantee that the count always equals the exact value

Answer: C) To release an approximate population count while limiting the influence of individual records

Explanation:

A differentially private count query adds calibrated randomness to the count. The result may differ from the true count, but its output distribution provides a formal privacy guarantee under the selected mechanism and parameters.

39. Why should a differentially private mechanism use an appropriate privacy accountant when making repeated releases?

  1. To increase the number of raw records available
  2. To track cumulative privacy loss and ensure releases remain within the chosen privacy limits
  3. To eliminate all noise from future queries
  4. To make every released statistic exact

Answer: B) To track cumulative privacy loss and ensure releases remain within the chosen privacy limits

Explanation:

Privacy accounting helps determine the overall guarantee when multiple mechanisms are composed. Without accounting, repeated queries or training steps can consume more privacy than intended, even if each individual operation has a seemingly small privacy cost.

40. What does the term privacy-utility trade-off mean?

  1. Increasing privacy always improves accuracy
  2. Privacy guarantees are unrelated to statistical quality
  3. Utility must always be zero when differential privacy is used
  4. Stronger privacy may require additional noise, which can reduce the usefulness or accuracy of results

Answer: D) Stronger privacy may require additional noise, which can reduce the usefulness or accuracy of results

Explanation:

Differential privacy balances limits on information leakage against the quality of released statistics or trained models. Selecting suitable privacy parameters and designing low-sensitivity queries can help maintain useful results while meeting privacy requirements.

41. Which approach can help reduce sensitivity for a numerical sum query?

  1. Restricting each individual's contribution to a known numerical range
  2. Allowing unlimited values for each individual
  3. Publishing all input values before computing the sum
  4. Disabling all input validation

Answer: A) Restricting each individual's contribution to a known numerical range

Explanation:

Bounding individual values or contributions limits how much the sum can change when one record changes. For example, restricting each contribution to a specified interval provides a finite sensitivity bound under an appropriately defined neighboring relation.

42. Which statement about differential privacy and public statistics is most accurate?

  1. Every published statistic must reveal each person's contribution
  2. Differential privacy can be used only for confidential financial data
  3. It can support publication of useful aggregate statistics with a formal bound on individual-level disclosure
  4. It guarantees that every public statistic is exactly correct

Answer: C) It can support publication of useful aggregate statistics with a formal bound on individual-level disclosure

Explanation:

Differential privacy is useful for publishing statistics such as counts, averages, and distributions. The mechanism introduces controlled randomness to limit the effect of any one individual's data, while retaining as much utility as practical.

43. What is the main purpose of the exponential mechanism?

  1. To encrypt a database using an exponential-time algorithm
  2. To select outputs from a discrete or general candidate set based on a utility function while satisfying differential privacy
  3. To guarantee that the highest-utility candidate is always selected
  4. To replace every statistical query with a password check

Answer: B) To select outputs from a discrete or general candidate set based on a utility function while satisfying differential privacy

Explanation:

The exponential mechanism selects an output with a probability influenced by its utility score and a privacy parameter. It is useful when the desired result is a choice from a set of candidates rather than a simple numerical answer.

44. Which statement best describes the role of delta in approximate differential privacy?

  1. It measures the exact number of users in the dataset
  2. It determines the number of columns in a table
  3. It always equals epsilon
  4. It provides an additive allowance in the privacy inequality and should be selected carefully

Answer: D) It provides an additive allowance in the privacy inequality and should be selected carefully

Explanation:

Delta is an additive parameter in approximate differential privacy. It should be chosen in accordance with the threat model, dataset scale, and privacy requirements rather than selected arbitrarily.

45. What is one benefit of using differential privacy in mobile analytics?

  1. It can help collect aggregate usage statistics while limiting what the collected outputs reveal about individual users
  2. It makes all mobile device data publicly accessible
  3. It removes the need for user consent where consent is legally required
  4. It guarantees that mobile applications cannot be compromised

Answer: A) It can help collect aggregate usage statistics while limiting what the collected outputs reveal about individual users

Explanation:

Differential privacy can help analyze trends such as feature usage or error frequencies without making released results overly dependent on any one user. The implementation must still account for repeated collection, privacy budgets, and applicable data protection requirements.

46. Which statement about the post-processing property is correct?

  1. It permits unlimited access to the original private dataset without further accounting
  2. It guarantees that every transformation improves statistical accuracy
  3. It means that data-independent processing of a differentially private output preserves its existing privacy guarantee
  4. It eliminates the need to define neighboring datasets

Answer: C) It means that data-independent processing of a differentially private output preserves its existing privacy guarantee

Explanation:

Once a result has been released under differential privacy, transformations that use only that result and independent information do not weaken the original privacy guarantee. A new computation that accesses private data must be analyzed separately.

47. A database administrator wants to publish an aggregate statistic without revealing the exact contribution of each individual. Which practice is most appropriate?

  1. Publish all individual records beside the statistic
  2. Apply a properly calibrated differentially private mechanism to the query
  3. Remove the database password but keep all raw records public
  4. Release the same exact individual-level information in a different file format

Answer: B) Apply a properly calibrated differentially private mechanism to the query

Explanation:

A differentially private mechanism adds randomness according to query sensitivity and the selected privacy parameters. This can enable the publication of useful aggregate results while limiting how much the output depends on any individual record.

48. Which statement about differential privacy guarantees is correct?

  1. It guarantees that an attacker can learn absolutely nothing about any individual
  2. It automatically protects all information stored on a compromised device
  3. It ensures every query result is perfectly accurate
  4. It bounds the change in output distributions between neighboring datasets according to specified privacy parameters

Answer: D) It bounds the change in output distributions between neighboring datasets according to specified privacy parameters

Explanation:

Differential privacy is a formal probabilistic guarantee, not a promise that no information can ever be inferred. It limits the additional influence of an individual's participation under a specified definition, while outputs may still reveal general patterns or information already present in the broader population.

49. A numerical query has an L1 sensitivity of 2. The Laplace mechanism is used with epsilon ε = 1. What is the scale parameter b of the added Laplace noise?

  1. 0.5
  2. 2
  3. 1
  4. 3

Answer: B) 2

Explanation:

For the standard Laplace mechanism, the scale is calculated as b = Δf/ε, where Δf is the L1 sensitivity. Substituting Δf = 2 and ε = 1 gives b = 2/1 = 2.

50. A differentially private system releases two results using mechanisms that each provide ε = 0.4 differential privacy. Under basic sequential composition, what is the combined epsilon guarantee?

  1. 0.2
  2. 0.4
  3. 0.8
  4. 1.6

Answer: C) 0.8

Explanation:

Under basic sequential composition, the epsilon values of two mechanisms are added. Therefore, the combined guarantee is ε = 0.4 + 0.4 = 0.8, assuming the mechanisms satisfy the stated individual guarantees and the composition applies to the releases.

Comments and Discussions!

Load comments ↻



Copyright © 2026 www.includehelp.com. All rights reserved.