Did the Model Understand the Question?
This work addresses the problem of model interpretability and robustness for researchers and practitioners in AI, revealing vulnerabilities in existing models through adversarial attacks.
The paper analyzed state-of-the-art deep learning models for question answering tasks and found that they often ignore important question terms, enabling the creation of adversarial examples that significantly drop accuracy, such as from 61.1% to 19% for visual question answering.
We analyze state-of-the-art deep learning models for three tasks: question answering on (1) images, (2) tables, and (3) passages of text. Using the notion of \emph{attribution} (word importance), we find that these deep networks often ignore important question terms. Leveraging such behavior, we perturb questions to craft a variety of adversarial examples. Our strongest attacks drop the accuracy of a visual question answering model from $61.1\%$ to $19\%$, and that of a tabular question answering model from $33.5\%$ to $3.3\%$. Additionally, we show how attributions can strengthen attacks proposed by Jia and Liang (2017) on paragraph comprehension models. Our results demonstrate that attributions can augment standard measures of accuracy and empower investigation of model performance. When a model is accurate but for the wrong reasons, attributions can surface erroneous logic in the model that indicates inadequacies in the test data.