A robust machine learning method for cell-load approximation in wireless networks

TL;DR

This paper introduces a robust machine learning algorithm for cell-load approximation in wireless networks, effectively handling uncertainty from limited training data in fast-changing environments.

Contribution

It develops a novel approximation framework that is robust, preserves key properties, and outperforms standard methods with small training sets.

Findings

Better robustness and accuracy with small training data

Preserves monotonicity and continuity properties

Outperforms standard approximation techniques

Abstract

We propose a learning algorithm for cell-load approximation in wireless networks. The proposed algorithm is robust in the sense that it is designed to cope with the uncertainty arising from a small number of training samples. This scenario is highly relevant in wireless networks where training has to be performed on short time scales because of a fast time-varying communication environment. The first part of this work studies the set of feasible rates and shows that this set is compact. We then prove that the mapping relating a feasible rate vector to the unique fixed point of the non-linear cell-load mapping is monotone and uniformly continuous. Utilizing these properties, we apply an approximation framework that achieves the best worst-case performance. Furthermore, the approximation preserves the monotonicity and continuity properties. Simulations show that the proposed method exhibits better robustness and accuracy for small training sets in comparison with standard approximation techniques for multivariate data.