AppleAnalysisPolicyNovember 11, 2026

Apple researchers propose SVT primitive to close federated DP utility gap

Read original source →machinelearning.apple.com

Apple paper shows any private selection algorithm built on standard noisy aggregation requires too many samples, then adds a Sparse Vector Technique primitive that matches central-model additive error when dimension is polynomially bounded. It also uses a near-optimal expected number of SVT queries, establishing an exponential separation from shuffle/aggregation models.

1 source

More stories today

Open the live feed