攻击神经网络
Fast Gradient Method (Explaining & Harnessing Adversarial Examples)
linear model:
where
理解: 在保持
DNN
where J is the loss function. In most cases (soft-max as classifier layer)
理解: 要让y’变化最大,及 应向loss 变大的方向变化。 即若
Iterative method
Basic iterative method
update equation (1)
Targeted
(3) becomes:
Our Method
What we have known and what we can obtained now?
Printable image size:
as the indexed boxes with size: Black/white box:Actual image (
) per frame. : starting index of the billboard for each frame. : range of the billboard for each frame. : gradient of each pixel in the bill board/
What we need to calculate
How?
Transform:
Nearest
SPP:
Binary
v1.5.2