List All Projects List Graduate Projects List Senior Design Projects List REU Projects Other Undergraduate Projects
Project # 72
Automating Container Security through Multi-Stage Vulnerability Scanning and Image Optimization
| Team Lead: | Ashwani Manoj Rai |
| Other Team Members: | Emmanuel OSEGHALE |
| Faculty Advisor(s): | Dr. Yuede Ji |
| Department: | CSE |
| Project Type: | REU |
| Title: | Automating Container Security through Multi-Stage Vulnerability Scanning and Image Optimization |
| Abstract: | High-Performance Computing (HPC) environments increasingly rely on containerized software stacks, yet vulnerability scanners provide only severity-centric assessments that do not reflect realistic exploitability within container runtime contexts. In this work, we propose a structured exploitability assessment pipeline that integrates vulnerability scanning, feature normalization, and large language model (LLM) reasoning to evaluate CVEs in container images under HPC conditions. We first collect a diverse set of HPC-oriented container images and identify vulnerabilities using industry-standard scanners. For each detected CVE, we extract and normalize structured features including CVSS metrics, EPSS scores, patch status, network exposure, authentication requirements, and container base characteristics (e.g., libc family). To mitigate label leakage and improve determinism, we enforce a strict feature schema and apply uncertainty-aware normalization. These structured features are then embedded into a standardized prompt and evaluated by a locally hosted LLM with multi-run aggregation and schema-constrained outputs. Our approach moves beyond severity-based ranking toward context-aware exploitability reasoning, enabling more realistic prioritization of vulnerabilities in HPC container deployments. |
| Poster: | View Poster #72 (PDF) |
| Judges Scoring: | Submit Evaluation (judges only) |