The Efficiency of Vulnerability Detection Based on Deep Learning

Computer software has been widely used in various industries, and many software cannot avoid receiving network attacks. The phenomena suggest that existing solutions for vulnerability detection demand improvement. This has motivated researchers to find and fix these vulnerabilities early. Deep learning is now widely used for vulnerability detection. Aiming at investigating the training efficiency of distinct neural network models, we leverage three datasets, covering 126 types of vulnerabilities. Each dataset is partitioned into three sets with a ratio of 6:2:2. Word2vec has been applied to transfer program code to vectors. Experiments results have shown that the DNN network achieved maximum efficiency.