5,321
edits
Deep Learning: Difference between revisions
→Fast Gradient Sign Method (FGSM)
| Line 668: | Line 668: | ||
\begin{aligned} | \begin{aligned} | ||
\delta_{FGSM} &= \max_{\Vert \delta \Vert \leq \epsilon} <\nabla l(f_{\theta}(x), y), \delta>\\ | \delta_{FGSM} &= \max_{\Vert \delta \Vert \leq \epsilon} <\nabla l(f_{\theta}(x), y), \delta>\\ | ||
&= \epsilon | &= \epsilon sign(\nabla l(f_{\theta}(x), y)) | ||
\end{aligned} | \end{aligned} | ||
</math> | </math> | ||