MLSecOps with Automated Online and Offline ML Model Evaluations on Kubernetes
summary
Open source tools run automated adversarial and fairness checks during training and at runtime using Adversarial Robustness Toolbox and AI Fairness 360, with integration into Kubernetes serving via KServe integration. A simple gradient attack drops MNIST accuracy from 87% to 13% in tests, the toolbox has >3,000 GitHub stars and 150,000+ downloads, online explainers run per request and offline logging via Kafka feeds periodic batch evaluations and Prometheus alerts.
AI-written summary · about
chapters (10)
- 0:00Start
- 11:07Financial Check Fraud
- 12:15Adversarial Threats to Al
- 13:02Real-world Adversarial Exploits
- 13:59Adversarial Threats to Machine Learning
- 20:46Pipeline: Adversarial Robustness Toolbox (ART) Component
- 24:05Online vs Offline model evaluations
- 26:19Inference Service Control Plane
- 28:44ART Explainer Spec
- 29:14ART Online Evaluation Demo Flow
related talks
40mMLSecOps with Automated Online and Offline ML Model Evaluations on Kubernetes
30mOwned by Statistics: Using Kubeflow to Defend vs Attacks on Your ML Models
42mMLOps Under Attack: Threat Modeling Modern AI Systems
53mSecurity of Machine Learning Systems
25mSecure ML: Automated Security Best Practices in Machine Learning
42mDevSecOps for AI: Introducing MLSecOps for Software 2.0
31mApplied MLOps to Maintain Model Freshness on Kubernetes
26mRunning Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline