Mitali Dhabekar completed Machine Learning for Automotive Safety

Built a perception validation dossier for a synthetic Aurora-7 detector end to end: defined detection using IoU ≥ 0.5, computed aggregate and VRU-specific metrics, exposed a severe night/occlusion recall collapse, and translated the top misses into ISO 21448 SOTIF chains with concrete mitigations. The final ship verdict was evidence-based and conditionally restricted to daylight/clear-visibility operation, with a real-world comparison to the Uber ATG Tempe crash that correctly mapped the failure to a night-time pedestrian miss.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Triage
• SOTIF Functional-Insufficiency Reasoning