FLEA: Provably Fair Multisource Learning from Unreliable Training Data

Published in Preprint, 2021

We propose a theoretically rigorous algorithm for detecting and eliminating malignant (perturbed) data in a multisource setup.’