ABSTRACT:

AI is everywhere. It is being applied to security as well. In our research on the security of medical and industrial IoT over the last 5 years, we have noticed several common mistakes, challenges, and issues in applying AI and securing IoT. In this talk, we will discuss nine such common issues and mistakes.

This talk covers the following topics:

• Overview
• AI-Based Security of IoT: Our Research
• Industrial Control Systems Security Using AI
• Internet of Medical Things Security Using AI
• Edge AI: Hierarchical Deep Learning
• Lessons Learnt: 10 Problems with AI Studies
• 1. No Domain Expertise
• 2. Random Datasets
• 3. Imbalance of Security Data
• 4. Wrong Metrics
• 5. Too Few or Too Many Features
• 6. Results Not Explainable
• 7. No Sensitivity Analysis
• 8. No Real-World Validation
• 9. Omitting Assumptions and Limitations
• Summary
• Our Publications
• Acronyms

Presentation slides in Adobe Acrobat Format: 1 slide/page

View Slides+Video

Right click and save to download Slides+Video podcast.