Automated Heuristic Design for Network Operations
For network operators, this work proposes a framework to automate heuristic design, potentially reducing reliance on expert knowledge, though results are preliminary and limited to a single task.
The paper explores applying Automated Heuristic Design (AHD) to network operations, demonstrating that AHD can generate heuristics for 5G Low-Density Parity Check decoding that match state-of-the-art production solutions.
Network operation relies on heuristics to solve many tasks rapidly and efficiently across the protocol stack. These heuristics are the result of thorough human-driven design rooted in expert knowledge of the target system and problem. Recently, approaches powered by Artificial Intelligence have shown promising results in devising solutions that outperform long-established heuristics in classical problems. We explore the possibility of applying such Automated Heuristic Design (AHD) frameworks to network environments by (i) discussing the general integration of AHD with network operation and the associated challenges, as well as (ii) proposing a practical implementation of AHD for a specific networking task, i.e., 5G decoding. Initial results show how modern AHD tools can devise heuristics for Low-Density Parity Check decoding on par with state-of-the-art solutions implemented in production systems.