DeepVuln

A deep learning based approach to detect vulnerability in C/C++ source code.

Process

• Pre-train a language model on a large corpus of C/C++ dataset
• Fine-tuning the model for the classification task of vulnerability detection in C/C++
• Evaluate the model on test dataset

Resources

Pre-training Dataset: C/C++ files from 100 GitHub repositories (Link)

Fine-tuning Dataset: Draper VDISC Dataset

Model: DistilBERT

Code: Project code is available on this colab notebook

Progress Tracker: Progress of the project can be tracked through this GitHub Project

References to Online Tutorials

• Pre-traing BERT
• Fine tunning