About ESV-Tracker
The first platform dedicated to identifying and classifying embedded system vulnerabilities, separating them from the noise of general-purpose CVEs.
Mission
Embedded systems power the most critical infrastructure around us, from medical devices and industrial controllers to IoT sensors and wireless sensor networks (WSN). Yet when it comes to vulnerability databases, embedded threats are buried alongside millions of general-purpose CVEs, making it nearly impossible for security teams to prioritize what truly matters.
ESV-Tracker was born from this gap. As a university research project developed at the Informatique & Applications Laboratory, our mission is simple: give embedded system security its own dedicated space, a platform that classifies, tracks, and surfaces vulnerabilities that specifically affect embedded and IoT environments.
How the Classification Works
At the core of ESV-Tracker is a Two-Stream Hybrid Embedding Architecture that combines two complementary models to classify each CVE as either an Embedded System Vulnerability (ESV) or a General-Purpose Software Vulnerability (GPSV).
The first stream uses Vuln2Vec, a custom domain-specific word embedding trained on vulnerability data, giving the model a deep understanding of specialized security terminology. The second stream leverages a large pre-trained language model (PLM) to capture broader contextual relationships and handle out-of-vocabulary (OOV) words that domain-specific models may miss.
The outputs of both streams are fused through a learnable weighted feature fusion mechanism, allowing the model to dynamically balance domain-specific precision with broad linguistic understanding — achieving state-of-the-art classification accuracy.
Data Sources & Roadmap
ESV-Tracker currently ingests and classifies vulnerabilities from the National Vulnerability Database (NVD), the most comprehensive publicly available CVE repository. Every entry is automatically processed through our classification pipeline upon ingestion.
Future data sources planned:
👥 Research Team RESI
📄 Published Work
The classification model powering ESV-Tracker is based on the following peer-reviewed publication: