Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore the critical challenges of building trustworthy AI systems for safety-critical applications in this 19-minute conference presentation from EDGE AI Milan 2025. Discover how imperceptible perturbations can cause AI models to miss crucial objects like cyclists, creating life-or-death scenarios in autonomous vehicles and other safety-critical systems. Learn comprehensive strategies for designing, verifying, and deploying robust AI systems that maintain reliability in real-world environments across transportation, industrial, and healthcare applications. Examine the "vanishing cyclist" problem and understand why small, nearly invisible changes to input data can dramatically alter AI decision-making in critical situations. Master techniques for building resilient AI pipelines through domain-specific labeling, synthetic data generation, and cross-platform interoperability across MATLAB, Python, PyTorch, TensorFlow, and ONNX frameworks. Delve into advanced explainability methods beyond traditional classification, including D-RISE techniques for object detection and image segmentation tasks. Understand formal verification approaches that provide mathematical guarantees for AI robustness within defined perturbation sets. Gain practical knowledge of edge deployment strategies including model compression, projection techniques, and automated code generation for CPUs, GPUs, and FPGAs. Implement runtime safeguards and out-of-distribution detection mechanisms to ensure AI system safety in unpredictable real-world scenarios. Connect engineering best practices to emerging regulatory frameworks like the EU AI Act and learn how to adapt traditional V-model development workflows for machine learning systems. Prepare your AI artifacts for regulatory audits and certification processes while maintaining system performance and reliability.
