ReNO-2: Supporting Human Network Operators with ML - A Cognitive Approach

Team

Distributed Systems, University of Kassel

  • Oliver Hohlfeld
    Oliver Hohlfeld is a Professor at the University of Kassel where he heads the Distributed Systems group. Before, he was professor at Brandenburg University of Technoligy and headded the Computer Networks group. He obtained his PhD from TU Berlin and was a member of Deutsche Telekom Innovation Laboratories. Oliver was a visiting scholar at the group of Paul Barford at the University of Wisconsin - Madison, USA.

Internet Network Architectures and Management, TU Berlin

  • Stefan Schmid
    Stefan Schmid is a Professor at the Technical University of Berlin, Germany. MSc and PhD at ETH Zurich, Postdoc at TU Munich and University of Paderborn, Senior Research Scientist at T-Labs in Berlin, Associate Professor at Aalborg University, Denmark, Full Professor at the University of Vienna, Austria, and Sabbathical as a Fellow at the Israel Institute for Advanced Studies (IIAS), Israel. Stefan Schmid received the IEEE Communications Society ITC Early Career Award 2016 and an ERC Consolidator Grant 2019.

  • Matthias Bentert
    PostDoc

Abstract

[EN] Many network outages today are caused by human errors. This project, ReNO-2 continues our ReNO-1 project of the first phase of the SPP 2378 program (Resilience in Connected Worlds), and aims to design resilient and adaptive networks even when the humans managing them introduce chaos. As in ReNO-1, we aim to support network operators in dealing with the complexity of network management and making networks more resilient using automation, leveraging tools from machine learning (ML) and formal methods (FM). The main focus of ReNO-2 is on the human operator, aiming to study and limit the effect of human errors on network resilience, also accounting for cognitive aspects. We hypothesize that a more cognitive approach is needed to support human operators work efficiently and accurately. Motivated by new technological opportunities and the combination of human-driven, ML-driven, and formal methods in network operations, and in particular the recent advent of Large Language Models (LLMs), we will explore key questions such as: How can we optimally support network operators with ML, LLM, and FM tools? How do ML and FM tools influence the confidence of network operators? Can we enhance LLM tools to provide more feedback, highlighting their limitations and risks? We will contribute methodologies to support human operators in using emerging ML, LLM, and FM tools and provide automatic feedback to users. Our methodologies target not only network operators but also university students who are learning about communication networks.

We will specifically consider two application scenarios, intra- and inter-domain networking. These are not only critical applications to keep the Internet connected but also have the advantage that formal method tools for configuration verification and synthesis (including our own from ReNO-1) exist. Additionally, in our labs, we have existing testbeds with which we can study how to support students optimally. The project is timely as the opportunities and risks of using ML and especially LLM networking tools are still poorly understood. Also, the essential cognitive aspects such as confidence (and how tools can influence it) have not yet received much attention in the literature.