search

danielzuegner / gnn-meta-attack

Implementation of the paper "Adversarial Attacks on Graph Neural Networks via Meta Learning".
https://www.kdd.in.tum.de/gnn-meta-attack
MIT License
143 stars 26 forks source link

Confusion about the Loss function L_{atk} #6

Closed harshakokel closed 4 years ago

harshakokel commented 4 years ago

Hello,

I am confused about the loss function in the paper Zuegner et al ICLR 2019.

In the paper, you mention that portfolio_view or portfolio_view and then gradient is

portfolio_view

But, in the algorithm the gradient equation is missing the -ve sign.

image

I am also confused about the same thing in the code. For 'Meta-Train' variant, I think the gradient calculation here is trying to minimize the classification error of the examples in the training here, instead of trying to maximize the classification error.

What am I missing?

danielzuegner commented 4 years ago

Hi, thanks for your message.

Sorry about the confusion. Since the attacker is trying to maximize the classification loss, Algorithm 1 is correct and in the paper it should read L_atk=L_self. Let me know if that helps!

Best,

Daniel

harshakokel commented 4 years ago

Thank you. This helps.