Definition of Privacy Parameters (ε and δ) and their importance
1. ε-Differential Privacy: The parameter ε (epsilon) is a non-negative
value that quantifies the privacy loss associated with a specific
computation. A mechanism is said to achieve ε-differential privacy if
the probability of obtaining any output is approximately the same
regardless of whether an individual’s data is included in the dataset or
not. A smaller ε indicates stronger privacy, implying that the impact of
any individual’s data on the analysis result is minimal.
2. δ-Differential Privacy: The parameter δ (delta) is another parameter
introduced to address the possibility of privacy breaches due to rare
events. It is used primarily in scenarios where the privacy guarantee
might be slightly relaxed for extremely small probabilities. A mechanism
achieves δ-differential privacy if the maximum allowable privacy loss is
bounded by δ.
While ε provides a clear measure of how differentially private an
analysis is, δ allows for fine-tuning the privacy guarantee to
accommodate rare situations while maintaining a strong level of privacy
overall.