Adversarial Robustness of Program Synthesis Models

Abstract

The resurgence of automatic program synthesis has been observed with the rise of deep learning. In this paper, we study the behaviour of the program synthesis model under adversarial settings. Our experiments suggest that these program synthesis models are prone to adversarial attacks. The proposed transformer model have higher adversarial performance than the current state-of-the-art program synthesis model. We specifically experiment with AlgoLisp DSL-based generative models and showcase the existence of significant dataset bias through different classes of adversarial examples.

Tasks

Reproductions