Context-Free Word Importance Scores for Attacking Neural Networks

Abstract

Leave-One-Outscores provide estimates of feature importance in neural networks for adversarial attacks. In this work, we present context-free word scores as a query-efficient alternative. Experiments show that these approximations are quite effective for black-box attacks on neural networks trained for text classification, particularly for CNNs. The model query count for this method scales as O(vocab_size *model_input_length). It is independent of the number of examples and features to be perturbed.