Targeted Universal Adversarial Perturbation Attack - GPU

[faresmalik]

Hello Team!

I am currently working on generating Targeted UAPs; I am running into a problem that could be faster! The attack is generated using CPU resources. However, I am specifying the device to be GPU, not CPU.

I am using the PyTorch classifier, and it takes so long to generate the UAPs.

When I delved into the code, it used np arrays rather than tensors.

Please provide any possible solution.