SPECTRE: Defending Against Backdoor Attacks Using Robust Statistics
Jonathan Hayase 1 Weihao Kong 1 Raghav Somani 1 Sewoong Oh 1
Abstract Starting with the seminal work of (Gu et al., 2017), there
Modern machine learning increasingly requires has been an active line of work on designing backdoor at-
training on a large collection of data from mul- tacks that use more stealth triggers (Chen et al., 2017; Liu
tiple sources, not all ...


雷达卡




京公网安备 11010802022788号







