首页▸人工智能与机器学习▸对抗性机器学习实验室 — 制作能欺骗分类器的扰动

⚡ 对抗性机器学习实验室 — 制作能欺骗分类器的扰动

将一个微小扰动拖到一个已被正确分类的点上,观察线性分类器如何翻转其判断——然后开启对抗训练,看看同样的攻击如何失效。

人工智能与机器学习2D中等60 FPS

⚙ Under the hood

Drag a small perturbation onto a correctly-classified point and watch a linear classifier flip its decision — then turn on adversarial training and watch the same attack stop working.

Canvas 2DAdversarial ExamplesFGSMRobustnessAdversarial Training

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install