Learning from End User Data with Shuffled Differential Privacy over Kernel Densities

Tal Wagner · 2025

Read Learning from End User Data with Shuffled Differential Privacy over Kernel Densities by Tal Wagner (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-14087.txt.

Description

Tal Wagner, a computer scientist, studies how to learn from data held on many user devices in a 2025 paper. Privacy is the core problem, since the raw data must stay safe while a model still learns from it. Wagner works in the shuffled model, where each user partly shields data before it is sent and the set is mixed so names are stripped. That model sits between full trust in one server and full local shielding, which harms accuracy.

The main result is a method to gauge how data points bunch up, known as a density, under this privacy model. Its accuracy comes close to what a trusted server could reach. Wagner uses it to train a sorting model, with one density learned per class. Tests show strong results, and the learned shapes still show what each class means.

Catalog record

CREATORTal Wagner
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-14087

Browse the full library · About the Tower of Babel