Anchal Munot completed Machine Learning for Automotive Safety

Built a complete perception-validation dossier for a synthetic Aurora-7 detector: defined detection at IoU ≥ 0.5, scored aggregate and VRU metrics, and showed that VRU false-negative rate—not headline accuracy—is the safety-critical measure. The candidate then sliced recall by condition, identified the night/occluded pedestrian collapse, wrote ISO 21448-style SOTIF chains with concrete mitigations, and issued a daylight-only conditional release recommendation grounded in the dossier’s own evidence.

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