Adversarial Examples for Evaluating Math Word Problem Solvers
This work addresses the robustness of MWP solvers for AI applications in education and reasoning, but it is incremental as it applies known adversarial techniques to a specific domain.
The paper tackled the problem of evaluating the true understanding of Math Word Problem (MWP) solvers by generating adversarial attacks, resulting in an average accuracy reduction of over 40 percentage points across three neural solvers on two benchmark datasets.
Standard accuracy metrics have shown that Math Word Problem (MWP) solvers have achieved high performance on benchmark datasets. However, the extent to which existing MWP solvers truly understand language and its relation with numbers is still unclear. In this paper, we generate adversarial attacks to evaluate the robustness of state-of-the-art MWP solvers. We propose two methods Question Reordering and Sentence Paraphrasing to generate adversarial attacks. We conduct experiments across three neural MWP solvers over two benchmark datasets. On average, our attack method is able to reduce the accuracy of MWP solvers by over 40 percentage points on these datasets. Our results demonstrate that existing MWP solvers are sensitive to linguistic variations in the problem text. We verify the validity and quality of generated adversarial examples through human evaluation.