The Formal Definition of Differential Privacy
What You'll Learn
You'll understand the precise mathematical guarantee that differential privacy provides: epsilon (ε) quantifies exactly how much plausible deniability each individual's data has in any analysis result.
The Core Mathematical Guarantee
Differential privacy has a rigorous formal definition that moves beyond intuition to provide a provable privacy guarantee. A randomized mechanism M satisfies ε-differential privacy if for any two datasets D₁ and D₂ that differ by exactly one record (one person's data), and for any possible output O:
Pr[M(D₁) = O] ≤ e^ε × Pr[M(D₂) = O]
This equation captures the essence: an observer cannot tell with confidence whether any individual's data was included or excluded from the dataset.
Understanding Epsilon (ε)
The parameter ε (epsilon) controls the privacy-utility tradeoff:
- ε = 0: Perfect privacy (but completely useless results—pure random noise)
- Small ε (e.g., 0.1-1.0): Strong privacy protection, individual records heavily obscured
- Large ε (e.g., 10+): Weak privacy, analysis results closer to true values
Think of ε as a "privacy budget": it bounds how much information can leak. Lower values mean individual records have better plausible deniability—whether you participated or not, the output distribution is nearly identical.
Why This Matters
Unlike vague promises like "anonymized data," ε-differential privacy provides a quantifiable, mathematically proven privacy level. An attacker with unlimited computing power and auxiliary knowledge still cannot determine if your record was included beyond the bound specified by ε. This makes privacy guarantees auditable and composable across multiple analyses.
Key Takeaway: ε-differential privacy mathematically guarantees that any single individual's data contribution is bounded by the factor e^ε, ensuring plausible deniability regardless of what else an attacker knows—making privacy measurable rather than aspirational.