Protecting sensitive datasets: cloud tools you can use

[2019-03-16 Sat 14:48:55] speaker: Felipa Hoffa

NYC taxi dataset

• sensitive data exposed

Cloud DLP (data loss preventation)

• redact sensitive data
• format preserving encryption

Data bucketing

k-anonymity

• at least k records have the same attributes
• k = 1 means record is unique among the sample
• usually require k=5

k-map, reidentification risk

• can zip code and age identify someone
• depends on the population of the zip code

L-diversity

• does everyone in the group have a common characteristic

delta-presence