Controlled Adversarial Text Generation

AbstractNLP models are shown to suffer from robustness issues, i.e., a model’s prediction can be easily changed under small perturbations to the input. In this work, we present a Controlled …...