Workshops Archive

SciNet: Science-Aware Network Operations for Research Networks


Workshop: 13th Annual International Workshop on Innovating the Network for Data Intensive Science — INDIS

Authors: Mariam Kiran (Oak Ridge National Laboratory (ORNL)); K. K. K. K. Ramakrishnan (University of California, Riverside); Harvey Newman (California Institute of Technology); Puneet Sharma (HPE); and Rafael Ferreira Da Silva (Oak Ridge National Laboratory)

Abstract: Emergence of AI-driven autonomous laboratories is fundamentally changing how scientific experiments are designed, executed, and analyzed. However, research networks remain passive communication infrastructures that seek to optimize packet delivery while being agnostic of the scientific workflows they support. This disconnect creates bottlenecks in distributed experimentation, multi-facility workflows, and real-time scientific decision making. In this paper, we propose Science-Aware Autonomous Networks (SciNet), a new networking paradigm where artificial intelligence (AI) is integrated into the network control plane to continuously observe scientific workflows, predict network demands, reason about workflow dependencies, and optimize network behavior, done autonomously. Unlike conventional AIOps systems that optimize utilization or reliability, SciNet is focused on optimizing for scientific productivity through workflow-aware routing, predictive resource allocation, and autonomous mitigation of failures. We present the architecture, mathematical formulation, learning framework, and evaluation methodology, and discuss how the proposed system aligns with the Department of Energy’s vision for AI-driven autonomous laboratories.


Back to 13th Annual International Workshop on Innovating the Network for Data Intensive Science — INDIS Archive Listing

Back to Full Workshop Archive Listing