A fuzzing test sample generation method based on GRU-A

After analyzing the current methods of fuzz testing, it can be concluded that there are issues such as low efficiency in generating test samples and low coverage. To address these shortcomings, a novel approach based on deep learning for generating fuzz testing samples is proposed. This method utilizes the GRU neural network and attention mechanism in deep learning to construct a model for generating fuzz testing samples. Through training the model, a set of semantic rules is learned, ensuring that the generated samples conform to semantic specifications. Finally, by crawling PDF object files, the two open-source PDF viewers are subjected to fuzz testing. The experimental results demonstrate that the proposed approach achieves higher coverage of fuzz testing samples and triggers a greater number of crashes.