20261106 - Sanjay Rao (Purdue) - "Principled AI approaches for automating network operations"

20261106 - Sanjay Rao (Purdue) - "Principled AI approaches for automating network operations" Authors: Sanjay Rao - Purdue Date: November 6th, 2026 Title: OpsSteward: Principled AI approaches for automating network operations Abstract: In this talk, I will present an overview of ongoing research in Purdue’s Internet Systems Laboratory on principled approaches to using AI to automate network operations spanning the management plane (agentic approaches for network configuration management, and root cause analysis), control plane (neural architectures for SDN controllers for routing and traffic engineering decisions), and the data plane (ML-based intrusion detection at line rates using programmable networking). I will go in depth on Elastispec, a recent system that we developed which audits enterprise firewall configuration files against thousands of natural language vendor documents. A key challenge of Elastispec is conducting formal verification despite dealing with the messy nature of enterprises (e.g., informal and partially documented specifications, incomplete and imperfect inventories). Evaluations with real-world enterprise firewall configurations and popular enterprise application vendor documents show Elastispec successfully identified 15 configuration errors in a production network. The Elastispec toolkit is publically available, and we would love campus operators to try it out and give us their feedback! More broadly, we would love to explore collaboration opportunities on any of our directions with network operators and other practitioners! For more information: https://engineering.purdue.edu/~sanjay/ https://purdue-isl.github.io/ https://github.com/Purdue-ISL/Elastispec