📜 [Paper] Edge-Case-Based AI Reliability Validation published in Scientific Reports (IF=4.9, Top 14.6%, with KTL)
🎉 Our paper, “Uncovering reliability blind spots in transportation AI: An edge-case-based validation approach for transportation systems,” has been published in Scientific Reports (Nature Portfolio, SCIE, IF = 4.9, JCR Top 14.6%).
This work is a collaboration with the Korea Testing Laboratory (KTL) — co-authored by June Young Kim (System & Energy Division, Industrial Convergence Technology Center, KTL) and Prof. Donghyun Lee (Corresponding). Congratulations! 🎊
Why this matters
Standard benchmark metrics can hide how an AI system behaves in the rare, difficult situations that matter most in high-risk domains. A model that looks nearly perfect on aggregate numbers may still fail systematically on the cases certification bodies care about.
What we did
Using vehicle license plate recognition as a case study, we extracted edge cases — challenging scenarios that baseline models misclassify — and used them as a supplementary validation set for AI reliability testing.
Key findings
- The conventional recall metric reported 99.2%, but evaluating the same system on the extracted edge cases yielded 93.3% — a substantial reliability gap invisible to standard evaluation.
- The proposed edge-case-based validation procedure is designed for certification bodies seeking to identify hidden vulnerabilities in AI systems before deployment.
Citation: Kim, J. Y., & Lee, D. (2026). Uncovering reliability blind spots in transportation AI: An edge-case-based validation approach for transportation systems. Scientific Reports. https://doi.org/10.1038/s41598-026-62679-w