| Title | An Empirical Study on Vulnerability Detection for Source Code Software based on Deep Learning |
| Publication Type | Conference Paper |
| Year of Publication | 2021 |
| Authors | Lin, Wei, Cai, Saihua |
| Conference Name | 2021 IEEE 21st International Conference on Software Quality, Reliability and Security Companion (QRS-C) |
| Date Published | dec |
| Keywords | codes, compositionality, Conferences, Deep Learning, Human Behavior, Logic gates, Manuals, Metrics, Neural Network, pubcrawl, Recurrent neural networks, Resiliency, software quality, vulnerability detection |
| Abstract | In recent years, the complexity of software vulnera-bilities has continued to increase. Manual vulnerability detection methods alone no longer meet the demand. With the rapid development of the deep learning, many neural network models have been widely applied to source code vulnerability detection. The variant of recurrent neural network (RNN), bidirectional Long Short-Term Memory (BiLSTM), has been a popular choice in vulnerability detection. However, is BiLSTM the most suitable choice? To answer this question, we conducted a series of experiments to investigate the effectiveness of different neural network models for source code vulnerability detection. The results shows that the variants of RNN, gated recurrent unit (GRU) and bidirectional GRU, are more capable of detecting source code fragments with mixed vulnerability types. And the concatenated convolutional neural network is more capable of detecting source code fragments of single vulnerability types. |
| DOI | 10.1109/QRS-C55045.2021.00173 |
| Citation Key | lin_empirical_2021 |
An Empirical Study on Vulnerability Detection for Source Code Software based on Deep Learning