EIFFeL: Ensuring Integrity for Federated Learning @GoogleTechTalks
EIFFeL: Ensuring Integrity for Federated Learning  @GoogleTechTalks
Uploaded May 2023 | Updated November 2026, 5 days ago
A Google TechTalk, presented by Amrita Roy Chowdhury (UC San Diego), 2023/04/19
ABSTRACT: Federated learning (FL) enables clients to collaborate with a server to train a machine learning model. To ensure privacy, the server performs secure aggregation of model updates from the clients. Unfortunately, this prevents verification of the well-formedness (integrity) of the updates as the updates are masked. Consequently, malformed updates designed to poison the model can be injected without detection. In this talk, I will formalize the problem of ensuring both update privacy and integrity in FL and present a new system, EIFFeL, that enables secure aggregation of verified updates. EIFFeL is a general framework that can enforce arbitrary integrity checks and remove malformed updates from the aggregate, without violating privacy. Further, EIFFeL is practical for real-world usage. For instance, with 100 clients and 10% poisoning, EIFFeL can train an MNIST classification model to the same accuracy as that of a non-poisoned federated learner in just 2.4s per iteration.
EIFFeL: Ensuring Integrity for Federated LearningLocal Node Differential Privacy2022 Blockly Developers Summit: Customizing BlocklyStreaming Attention Approximation via Discrepancy TheoryDifferentially Private Diffusion Models Generate Useful Synthetic ImagesDifferentially Private Diffusion ModelsNASA CARA - Air Traffic Control in SpaaaaaaaaceWorst-Case Membership Inference of Language ModelsStable Estimators for Fast Private Statistics2022 Blockly Developers Summit: Contributing to BlocklyCharles Hoskinson | CEO of Input Output Global | web3 talks | Apr 4th 2024 | MC: Marlon Ruiz2022 Blockly Developers Summit: Backwards Execution
Google TechTalks |

EIFFeL: Ensuring Integrity for Federated Learning

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER