Securing AI Models Training in Thimphu, Bhutan

This training provides hands-on and theoretical knowledge for identifying and defending against security threats and attacks targeting AI/ML models. It covers vulnerabilities in model development, training, and deployment pipelines, and equips participants with practical tools and defensive strategies to safeguard AI systems across use cases.

Understanding AI Threats and Attack Vectors

Overview of AI Model Security

• What makes AI models vulnerable
• AI security vs traditional software security
• AI attack taxonomy: training time vs inference time

Data Poisoning and Training Attacks

• Poisoned inputs and backdoors
• Label flipping and data injection

Adversarial Examples and Inference-Time Attacks

• How small changes can fool AI
• Image, text, and audio adversarial attacks

Discussion and Case Studies

• Real-world examples (e.g., autonomous driving, facial recognition)
• Failures due to adversarial or poisoned data

Model Theft, Privacy, and Defense Mechanisms

Model Extraction and Inversion Attacks

• API-based model stealing
• Membership inference and model inversion
• Risks of exposing model outputs and confidence scores

Defense Strategies – Part 1

• Adversarial training and gradient masking
• Model ensembling and input preprocessing
• Hands-on: Apply basic defenses to a model and test robustness

Privacy-Preserving Machine Learning

• Differential privacy
• Federated learning and secure aggregation
• Use of tools like PySyft and Opacus for privacy

Secure Deployment and Advanced Defenses

Secure Model Deployment and Access Control

• Threats in model hosting and serving environments
• API throttling, output sanitization, and rate limiting
• Securing endpoints and pipelines

AI Red Teaming and Monitoring Tools

• Overview of red teaming AI models
• Toolkits: IBM Adversarial Robustness Toolbox (ART), CleverHans, Foolbox
• Monitoring AI systems for anomalies

Wrap-Up and Review

• Review of tools, techniques, and best practices
• Discussion on evolving threats and trends
• Q&A, resources, and course feedback