Distributed Adversarial Examples Generation #1229
Description
Hi,
The adversarial examples generation is always time-consuming in adversarial training. Is there a way to conduct multi-GPU generation in tensorflow2?
e.g. For a batch with 128 samples using 4 GPUs, can I generate 128/4 adversarial examples by each GPU at the same time and concatenate them later?
Look forward to hearing from you.