Index Ai Academy
Glossary · Phase 11 · Professional practice

Membership inference

Also written: membership inference attack

Working out whether a particular record was in a model's training data, usually from how confident the model is about it.

A trained model is made from its training data, and can give some of it back. One measurable form of this is a membership inference attack: given a person's record, decide whether the model was trained on it. That alone can be the secret, if the training set was "patients with this condition".

Try it now

Each row keeps fewer details than the one above. k is the size of the smallest group.

Learn it properly

Privacy engineering · Why removing names isn't anonymity, how to hold less (minimization, retention, keyed pseudonyms), the idea behind differential privacy and its budget, what a model remembers, and a deletion that reaches every store. About 40 minutes.

Other terms from the same lesson

  • data minimization: Collecting and keeping only the data a feature needs, for only as long as it needs it.
  • differential privacy: Adding random noise to answers so that they look much the same whether or not any one person is in the data. A number called epsilon sets how much.
  • direct identifier: A column that names one person by itself, such as an email address, a phone number or an account id.
  • privacy budget: The total amount of privacy a system will spend on noisy answers. Each answer uses some, and when it is gone no more are given.
  • pseudonym: A stand-in for an identifier, such as a keyed hash of an email address. Whoever holds the key can link it back, so the data is still personal data.
  • quasi-identifier: A column that doesn't name a person by itself, such as a ZIP code or birth date, and can single one out in combination with others.
Keep learning

More in plain English.

1286 terms in the glossary, each with the lesson that teaches it.

1286 terms

The glossary

Every word you'll meet in AI and machine learning, explained in everyday language.

Phase 11 · lesson 40 min

Privacy engineering

Why removing names isn't anonymity, how to hold less (minimization, retention, keyed pseudonyms), the idea behind differential privacy and its budget, what a model remembers, and a deletion that reaches every store.

Never trained an AI? Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account. Teach it to write