Course contentsShow
← Security
Lesson 2915 of 3,05254. Privacy EngineeringFree lesson

The Laplace Mechanism

Adding calibrated noise to numeric queries using the Laplace distribution to achieve differential privacy guarantees.

The Laplace Mechanism

What You'll Learn

You'll learn how to add carefully calibrated noise to numeric query results using the Laplace distribution to satisfy differential privacy guarantees while preserving data utility.

The Core Idea

Now that you understand the privacy budget (epsilon) and the formal differential privacy guarantee, you need a practical method to achieve it for numeric queries. The Laplace Mechanism is the foundational technique: it adds random noise drawn from a Laplace distribution to query outputs, where the noise scale is calibrated to the query's sensitivity and your chosen epsilon.

How It Works

The mechanism follows three steps:

  1. Compute the true answer to your query (e.g., average age, count of records)
  2. Determine sensitivity: How much could one person's data change the result? (This is the query's global sensitivity)
  3. Add Laplace noise scaled by sensitivity/epsilon

The Laplace distribution is symmetric and centered at zero, with a "scale" parameter that controls spread. Larger epsilon (weaker privacy) means less noise; smaller epsilon (stronger privacy) means more noise.

Analogy

Imagine a thermometer that always gives the true temperature plus or minus a random amount from a specific bell-shaped curve. The "width" of that curve depends on how private you want readings to be. A narrow curve (high epsilon) keeps readings close to truth; a wide curve (low epsilon) obscures individual contributions but protects privacy.

Why Laplace?

The Laplace distribution's exponential tails perfectly match the differential privacy definition's multiplicative guarantee. Adding Laplace noise with scale Δf/ε (where Δf is sensitivity) provably satisfies ε-differential privacy.

Key Takeaway:
The Laplace Mechanism achieves differential privacy by adding calibrated random noise from a Laplace distribution, with noise magnitude controlled by the ratio of query sensitivity to privacy budget (epsilon).