InSlicing: Interpretable Learning-Assisted Network Slice Configuration in Open Radio Access Networks

TL;DR

This work tackles dynamic, non-convex network-slice configuration in open RANs by introducing InSlicing, an interpretable framework that first learns per-slice performance mappings $P(X)$ with Kolmogorov-Arnold Networks and then solves the transformed optimization problem via a genetic algorithm enhanced with a trust-region local refinement. The approach yields high interpretability through edge-based activations in KANs and achieves substantial operation-cost reductions (over 25%) compared with strong baselines, validated on a realistic OpenAirInterface testbed with 9 slices. Key contributions include the use of KANs for transparent performance modeling, a hybrid GA-TRM optimization workflow, and demonstrations of scalability, low regret, and precise SLA compliance. The practical impact lies in enabling deployable, explainable, cost-efficient slice configurations for next-generation open RAN deployments. All mathematical notation is presented with explicit delimiters, e.g., $P(X)$, $X_i$, and $C(w_r, x_{i,r})$.

Abstract

Network slicing is a key technology enabling the flexibility and efficiency of 5G networks, offering customized services for diverse applications. However, existing methods face challenges in adapting to dynamic network environments and lack interpretability in performance models. In this paper, we propose a novel interpretable network slice configuration algorithm (\emph{InSlicing}) in open radio access networks, by integrating Kolmogorov-Arnold Networks (KANs) and hybrid optimization process. On the one hand, we use KANs to approximate and learn the unknown performance function of individual slices, which converts the blackbox optimization problem. On the other hand, we solve the converted problem with a genetic method for global search and incorporate a trust region for gradient-based local refinement. With the extensive evaluation, we show that our proposed algorithm achieves high interpretability while reducing 25+\% operation cost than existing solutions.

Figures (7)

• Figure 1: The overview of the InSlicing algorithm.
• Figure 2: The learning loss and illustration of KANs.
• Figure 3: The heatmap of slice configuration in InSlicing.
• Figure 4: The cost under comparison algorithms.
• Figure 5: The scalability under comparison algorithms.