AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

Private Federated Learning with Autotuned Compression

Enayat Ullah, Christopher A. Choquette-Choo, Peter Kairouz, Sewoong Oh

ICML 2023 · Federated learning and privacy › Differential privacy · approach: Private online and user-level learning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Adapt Norm with c0=0.1 (automatic sketch size) instead of grid-searched Genie compressionAdapt Norm FL · F-EMNIST, Shakespeare, SONWPvalidation accuracy and compression rateMixed
Raise relative compression error constant c0 from 0.1 to 0.25 (ablation)Adapt Norm FL · F-EMNIST, z=0.1compression rate and validation accuracy: accuracy 81.34 -> 78.87; compression 47 -> 101Mixed
Lower relative compression error constant c0 from 0.1 to 0.01 (ablation)Adapt Norm FL · F-EMNIST, z=0.1compression rate and validation accuracy: accuracy 81.34 -> 82.51; compression 47 -> 5Mixed
Two Stage method (W=75 uncompressed warm-up rounds, then fixed compression) instead of Adapt Norm (ablation)Two Stage FL · F-EMNIST, Shakespeare, SONWPaverage compression rateHurt
Adapt Tail (adaptive over log d rounds) used in FLAdapt Tail FL · federated learning benchmarksmodel accuracyHurt
Adaptive sketch compression (Adapt Norm) vs non-adaptive sketching on F-EMNIST, runtime per roundAdapt Norm FL · F-EMNISTseconds per round: 3.63s -> 3.69s per roundNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy