Adversarial Robustness of Visual Dialog
This work addresses safety and reliability concerns for multimodal AI systems in dialog applications, though it is incremental as it extends adversarial robustness research to a new domain.
This study investigates the adversarial robustness of visually grounded dialog models against textual attacks, finding that models encoding dialog history are more robust and that both textual and visual contexts are crucial for generating plausible adversarial examples.
Adversarial robustness evaluates the worst-case performance scenario of a machine learning model to ensure its safety and reliability. This study is the first to investigate the robustness of visually grounded dialog models towards textual attacks. These attacks represent a worst-case scenario where the input question contains a synonym which causes the previously correct model to return a wrong answer. Using this scenario, we first aim to understand how multimodal input components contribute to model robustness. Our results show that models which encode dialog history are more robust, and when launching an attack on history, model prediction becomes more uncertain. This is in contrast to prior work which finds that dialog history is negligible for model performance on this task. We also evaluate how to generate adversarial test examples which successfully fool the model but remain undetected by the user/software designer. We find that the textual, as well as the visual context are important to generate plausible worst-case scenarios.