Danyl Bince completed Machine Learning for Automotive Safety

Built a perception-validation dossier that defined detection rigorously with an IoU threshold, computed precision/recall plus the VRU false-negative rate from a scored synthetic table, and exposed how performance collapsed in night and occluded VRU slices. The candidate then translated those misses into ISO 21448 SOTIF chains with concrete mitigations and tied the analysis back to the real Uber ATG Tempe crash using the dossier’s own evidence.

• Leakage-Free Detection Definition
• Honest Metric Evaluation
• Per-Condition Safety Triage
• SOTIF Functional-Insufficiency Reasoning