Differential Privacy Implementation

This project demonstrates the Laplace and Exponential Mechanisms for differential privacy using an adult dataset.

Laplace Mechanism

• Global Sensitivity: Calculated as (max age - min age) / record count for ages above 25.
• Steps:
  1. Compute average age for ages > 25.
  2. Determine global sensitivity.
  3. Calculate variance using sensitivity and epsilon.
  4. Add Laplace noise to the average age.
  5. Output the noise-added result.

Higher epsilon values yield more accurate results; lower values increase privacy.

Exponential Mechanism

• Global Sensitivity: 1, as adding/removing a record changes one education level.
• Steps:
  1. Calculate utility scores for education levels.
  2. Use a scaling factor for large counts.
  3. Compute probabilities with epsilon and sensitivity.
  4. Output based on probabilities.

Lower epsilon values increase privacy, altering the most frequent education output.

Conclusion

This project illustrates the balance between privacy and accuracy in differential privacy.