Universal Adversarial Triggers for Attacking and Analyzing NLP
This work addresses security and interpretability issues in NLP models for researchers and practitioners, offering a novel attack method and analysis tool.
The paper tackles the problem of model vulnerabilities in NLP by introducing universal adversarial triggers, input-agnostic token sequences that cause models to make specific predictions when added to any input, resulting in significant drops in accuracy such as from 89.94% to 0.55% on SNLI and triggering harmful outputs like racist text from GPT-2.
Adversarial examples highlight model vulnerabilities and are useful for evaluation and interpretation. We define universal adversarial triggers: input-agnostic sequences of tokens that trigger a model to produce a specific prediction when concatenated to any input from a dataset. We propose a gradient-guided search over tokens which finds short trigger sequences (e.g., one word for classification and four words for language modeling) that successfully trigger the target prediction. For example, triggers cause SNLI entailment accuracy to drop from 89.94% to 0.55%, 72% of "why" questions in SQuAD to be answered "to kill american people", and the GPT-2 language model to spew racist output even when conditioned on non-racial contexts. Furthermore, although the triggers are optimized using white-box access to a specific model, they transfer to other models for all tasks we consider. Finally, since triggers are input-agnostic, they provide an analysis of global model behavior. For instance, they confirm that SNLI models exploit dataset biases and help to diagnose heuristics learned by reading comprehension models.