We'd like to apply expectation over transformation (EoT - https://proceedings.mlr.press/v80/athalye18b/athalye18b.pdf) to learn an adversarial chunk that is robust to perturbations, such as random deletion of one of its tokens.
This should be doable without modifying the core library.
We'd like to apply expectation over transformation (EoT - https://proceedings.mlr.press/v80/athalye18b/athalye18b.pdf) to learn an adversarial chunk that is robust to perturbations, such as random deletion of one of its tokens.
This should be doable without modifying the core library.