Ashmit Kumar completed Machine Learning for Automotive Safety

Built a perception validation dossier that correctly defined detection by IoU threshold, scored synthetic detections with precision/recall and VRU false-negative rate, and exposed the safety-critical collapse in night and occluded VRU recall. The work ties those numbers into a condition-based triage and a deployment recommendation grounded in the real Uber ATG Tempe crash.

• Leakage-Free Detection Metric Framing
• Condition-Sliced Safety Analysis
• Safety-Critical Metric Prioritization