From Edge To Insight: Cybersecurity Intelligence With Kubeflow...
summary
Proof of concept architecture ingests Falco logs at the edge, runs a GNN over cluster events runs GNN inference under 100 ms and scores anomalies locally, exports models to ONNX so they ONNX-exported models run on CPU and can be deployable to K3s edge clusters across ~100 locations; central Kubeflow pipelines (Katib, KFServing, JupyterHub, MinIO, Harbor, Keycloak/Dex) orchestrate a GraphRAG with vLLM + knowledge graph aligned to MITRE/NIST, enrich signals, and push automated policy/rule updates back to the edge — POC works end-to-end.
AI-written summary · about
chapters (12)
- 0:00Cybersecurity and AI defense intro
- 3:18Dynamic threat detection graph
- 7:17ML lifecycle management challenges
- 9:08Kubeflow edge-cloud architecture
- 13:43Four-tier ML pipeline and serving
- 16:32Scaling lightweight GNN deployment
- 19:19Automating ML with Kubeflow Pipelines
- 20:25Edge-cloud continuous learning loop
- 22:15Real-time anomaly defense
- 26:08Demo: automation and authentication
- 29:17Key lessons and roadmap
- 30:50Q&A
related talks
36mShift-Left for LLMs: Securing the AI Model Supply Chain
29mA Scalable Platform for Training and Inference Using Kubeflow
35mAdvanced Model Inferencing Leveraging KNative, Istio & Kubeflow Serving
30mOwned by Statistics: Using Kubeflow to Defend vs Attacks on Your ML Models
31mKubeflow - Data Science on Steroids
27mKubeflow for enabling AI Powered Drug Discovery and development...
25mSecuring AI Pipelines: Real-World Attacks on Kubernetes-Based AI Infrastructure
14mLightning Talk: A GNN Based Framework for Kubernetes Security Agents: Threat & Vulnerability...