Explore PII detection, data anonymization
techniques,
and differential privacy.
search Input Text
warning Detected
PII
shield Redacted Output
Hiding in a crowd of k
Each sheet is one patient. Patients whose generalised age and ZIP match are stacked: they
can no longer be told apart, so an attacker who knows your age and ZIP is left guessing
among everyone in your stack. k-anonymity asks that the smallest stack hold at least k.
Raising k merges the stacks but blurs the data: watch the information loss climb. Try it on
paper first with the lecture worksheet, then set k below and compare.
The tool's own lattice search on its 36-row table, suppression budget 20%.
Dashed sheets: records suppressed rather than generalised.
table_chart Sample
Dataset
Name
Age
ZIP
Condition
Income
Quasi-identifiers: Age, ZIP ·
Sensitive: Condition
The tool searches the generalization lattice for the least-damaging Age and ZIP
generalization that meets your constraints, suppressing outlier records rather than
over-generalizing everyone.
l = 1 means k-anonymity only. l-diversity is an
extra constraint on top of k-anonymity, not a substitute for it.
verified_user
Anonymized Result
Name
Age
ZIP
Condition
Income
tune Parameters
Differential privacy adds noise calibrated to a
query's global sensitivity Δf. Laplace draws from Lap(Δf/ε) and gives pure ε-DP. Gaussian
draws from N(0, σ²) with σ = Δf·√(2·ln(1.25/δ))/ε and gives the weaker (ε, δ)-DP.
Every release costs budget. Releasing n noisy
answers costs n·ε, not ε — and running both mechanisms on the same data costs both. This
panel charges you for all of it, which is why the budget moves faster than you might expect.