How to customize the loss function

[xmser]

First of all, thank you for your great contribution. This is really an amazing result.
I need to use adversarial training (I just touched on this issue ) in my research topic. I want to do a robust binary classification problem using logistic regression, but I am not very sure how to rewrite the loss function. For example, how to customize the loss function in the picture .Looking forward to your reply. Thank you very much.