Scaling Machine Learning Without Compromising Privacy
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 13:18· Intrado Virtual Event Platform,·
summary
Claims a concrete stack for locking down distributed ML pipelines at scale: SPIRE/SPIFFE identities for transient service and user identities, OPA policy enforcement for fine-grained admission and runtime controls, and Envoy mTLS sidecars to enforce service-to-service authentication and network isolation. Integrates LDAP/OIDC for user auth, admission controllers to restrict image registries, and explores SPIFFE federation for on‑prem/cloud trust domains. Example target is a multi-cluster financial ML deployment; the approach reduces exposed attack surface while preserving pipeline distribution.
AI-written summary · about
chapters (5)
related talks
19mFederate, Scale, and Secure: Practical SPIFFE/SPIRE for Containers...
12mIAM and Privileged Access Foundations for Secure ML Platforms
34mSecuring Multi-Cloud Cross-Cluster Communication with SPIFFE and SPIRE
32mA Platform-Agnostic Approach to Securing Workloads
37mLeveraging Cryptographic Lineage for Context and ConnectedTrust
31mScaling privacy embedded distributed ML systems
19mSecuring AI Pipelines from Development to Production
33mSPIFFE and SPIRE: Architecture Deep Dive