Sensibilité locale is a concept in confidentialité des données and apprentissage automatique that refers to the degree to which a small change in an individual’s input data can affect the output of a function or algorithm. It is particularly relevant in the context of ensuring the privacy of sensitive information while still allowing for useful analysis.
In technical terms, local sensitivity is often defined as the maximum change in the output of a function when a single individual’s input is altered. This is crucial in applications such as confidentialité différentielle, where the goal is to provide accurate results while preventing the identification of any particular individual’s data.
For example, consider a function that calculates the average income of a group of individuals. If the income of one person changes slightly, the average income should ideally not change significantly. The local sensitivity of this function would help quantify how much the average would shift in response to that individual’s income change.
Local sensitivity can also be used to design algorithms that add noise to outputs, effectively masking the influence of any single individual’s data. By understanding and calculating local sensitivity, researchers and data scientists can create more robust privacy-preserving mechanisms that maintenir l’utilité des données tout en protégeant la vie privée individuelle.
Dans l’ensemble, la sensibilité locale est un concept essentiel à l’intersection de analyse de données and privacy, enabling responsible handling of sensitive information in an increasingly data-driven world.