FAKE IT TILL YOU MAKE IT
date
Jan 1, 2024
Last edited time
Jan 1, 2024 03:25 PM
status
Published
slug
FAKE IT TILL YOU MAKE IT论文阅读
tags
summary
type
Post
origin
Field
Plat
(ICLR2023)FAKE IT TILL YOU MAKE IT: TOWARDS ACCURATE NEAR-DISTRIBUTION NOVELTY DETECTION
我们的目标是进行基于图像的新颖性检测。尽管取得了相当大的进展,但现有模型要么失败,要么在所谓的“接近分布”设置下面临戏剧性的下降,即正常样本和异常样本之间的差异微妙。
我们提出使用基于得分的生成模型来产生合成的近似分布的异常数据。然后对我们的模型进行微调,以区分这些数据与正常样本。
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2F99debced-a214-4be1-8be6-6b2bd51e9cb9%2FUntitled.png?table=block&id=9d6a38df-c35e-45c2-a7db-55d811819079&cache=v2)
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2F908fd4f8-aef5-418d-80f9-c0d04db8319b%2FUntitled.png?table=block&id=425e8f2b-dc66-4bd4-bf2d-4a1ff39e02dd&cache=v2)
Ablation
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2F33e0af1a-4107-4488-a6f7-03f87b21604a%2FUntitled.png?table=block&id=07e12e25-7dbc-44b2-a39b-4cb7e212923d&cache=v2)
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2Fe021c281-0078-4dfa-8283-01382457e2ec%2FUntitled.png?table=block&id=2ba5c129-b295-45f1-8075-8c764f249045&cache=v2)
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2Fad8f4eaf-f6ec-43d6-a5a1-31f6e311029c%2FUntitled.png?table=block&id=4e63e3e9-0d8d-4b1a-98a5-015a3745a99c&cache=v2)
![notion image](https://www.notion.so/image/https%3A%2F%2Fprod-files-secure.s3.us-west-2.amazonaws.com%2Fd919c123-ae4b-49b3-af3c-0184fe33faac%2F1aa53f32-97e9-4218-8292-94cd999ec464%2FUntitled.png?table=block&id=5f0ba61a-5553-4500-8fc6-144a22668d47&cache=v2)