When AI can’t see the cyclist… trust becomes the real challenge
A cyclist disappears to the model—but not to your eyes. That mismatch is the heart of safety-critical AI.
In this presentation from EDGE AI Milan 2025, Lucas Garcia, Product Manager at MathWorks, explores how to design, verify, and deploy trustworthy AI systems that hold up in real-world, safety-critical environments—from cars and planes to metros and medical devices.
💡 Inside the talk:
• The “vanishing cyclist” problem—and why imperceptible perturbations can flip life-or-death decisions
• Building resilient pipelines: domain-specific labeling, synthetic data, and cross-platform interoperability across MATLAB, Python, PyTorch, TensorFlow, and ONNX
• Explainability beyond classification with D-RISE for detectors and segmentation
• Formal verification for robustness—mathematical guarantees within defined perturbation sets
• Edge deployment with model compression, projection, and code generation for CPUs, GPUs, and FPGAs
• Runtime safeguards and out-of-distribution detection for real-world safety
Throughout the session, Lucas connects this engineering foundation to emerging standards like the EU AI Act and evolving workflows that adapt the V-model for learning systems, ensuring your artifacts are ready for audits and certification.
This is a must-watch for engineers, data scientists, and safety professionals building trustworthy edge AI in transportation, industrial, and healthcare systems.
🎥 Watch the full presentation → https://youtu.be/LNikcPxqDQM
🌐 Learn more → https://edgeaifoundation.org
#edgeAI #MathWorks #TrustworthyAI #AIForSafety #EmbeddedAI #OnDeviceAI #EdgeComputing #AIExplainability #AIStandards #AIInnovation #AIVerification #TinyML #EUAIAct #AICommunity #AIEvents #AIHardware
source
