“I'm an AI researcher working on generative modeling, so I want to note my bias upfront. But I'm also a musician/artist and so I find some of these conversations odd. The problem with them I see is that they are oversimplified. To get better at drawing I often copy other works. Or I'll play a piece exactly as intended. Then I get more advanced and learn a…”
This Person (Probably) Exists. Identity Membership Attacks Against GAN Generated Faces
Recently, generative adversarial networks (GANs) have achieved stunning realism, fooling even human observers. Indeed, the popular tongue-in-cheek website {\small \url{ http://thispersondoesnotexist.com}}, taunts users with GAN generated images that seem too real to believe. On the other hand, GANs do leak information about their training data, as evidenced by membership attacks recently demonstrated in the literature. In this work, we challenge the assumption that GAN faces really are novel creations, by constructing a successful membership attack of a new kind. Unlike previous works, our attack can accurately discern samples sharing the same identity as training samples without being the same samples. We demonstrate the interest of our attack across several popular face datasets and GAN training procedures. Notably, we show that even in the presence of significant dataset diversity, an over represented person can pose a privacy concern.
候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及- This Person (Probably) Exists. Identity Attacks Against GAN Generated Faces ↗sohkamyung · 2021-10-14 · 2 分