Date of Award

6-8-2026

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2026

Degree Name

Master of Science (MS)

Department

Computer Science and Engineering

First Advisor

Sean Choi

Abstract

Despite decades of progress in network protocols and control algorithms, many network-control mechanisms continue to rely on fixed heuristics, manually written rules, or static controller logic. This thesis explores the use of large language models as reasoning-based decision-making components within networked systems.

The thesis presents two case studies. The first, RLVR-CC, investigates whether a large language model can be trained to make congestion-control decisions using reinforcement learning with verifiable rewards. In this system, the model observes network state, reasons about current conditions, and selects a rate-control action. The second, Agent-Defined Networking, studies whether a large language model can participate in software-defined networking control-plane decisions, including natural-language policy compilation, runtime firewall enforcement, and congestion-aware routing.

Together, these systems examine the potential and limitations of incorporating large language models into network infrastructure. The results suggest that language models can support network decision-making when their outputs are bounded by reward functions, validation checks, and safety guardrails.

Share

COinS