Software Vulnerability Detection

Cybersecurity is one of those fields where the stakes are incredibly high - one vulnerability can compromise millions of users. I wanted to see if we could use deep learning to automatically detect vulnerabilities in code before they become problems. We developed MDSADNet, which is a multi-scale convolutional neural network that looks at code at different levels of abstraction. The results were pretty exciting - we hit a 98% F1-score, which means the model is both precise and comprehensive in finding vulnerabilities. This research showed me how AI can be a powerful tool for making software more secure, essentially giving developers an extra set of eyes that never gets tired.