Attack ML Models
摘要：
Attack Training的Loss:固定x,修改θ,使y0接近ytrue. Non-targeted Attack的Loss:固定θ,修改x,使y‘远离ytrue. Targeted Attack的Loss:固定θ,修改x,使y‘远离ytrue且接近yfalse. 4个pixel,(RGB每个 阅读全文
posted @ 2020-02-26 10:53 cltt 阅读(259) 评论(0) 推荐(0) 编辑
