Multi-Generator Continual Learning for Robust Delay Prediction in 6G

TL;DR

The paper tackles maintaining accurate one-way delay predictions in 6G networks under distributional shifts via continual learning. It introduces a multi-generator generative replay framework that uses tabular variational autoencoders and a domain-guided generator selector to mitigate catastrophic forgetting. The approach leverages UE-device-type modalities to allocate replay generators and improve tail performance, validated on a realistic 5G testbed against baselines. Results show superior stability, tail accuracy, and reduced data storage needs, highlighting practical benefits for proactive network management.

Abstract

In future 6G networks, dependable networks will enable telecommunication services such as remote control of robots or vehicles with strict requirements on end-to-end network performance in terms of delay, delay variation, tail distributions, and throughput. With respect to such networks, it is paramount to be able to determine what performance level the network segment can guarantee at a given point in time. One promising approach is to use predictive models trained using machine learning (ML). Predicting performance metrics such as one-way delay (OWD), in a timely manner, provides valuable insights for the network, user equipments (UEs), and applications to address performance trends, deviations, and violations. Over the course of time, a dynamic network environment results in distributional shifts, which causes catastrophic forgetting and drop of ML model performance. In continual learning (CL), the model aims to achieve a balance between stability and plasticity, enabling new information to be learned while preserving previously learned knowledge. In this paper, we target on the challenges of catastrophic forgetting of OWD prediction model. We propose a novel approach which introducing the concept of multi-generator for the state-of-the-art CL generative replay framework, along with tabular variational autoencoders (TVAE) as generators. The domain knowledge of UE capabilities is incorporated into the learning process for determining generator setup and relevance. The proposed approach is evaluated across a diverse set of scenarios with data that is collected in a realistic 5G testbed, demonstrating its outstanding performance in comparison to baselines.

Figures (8)

• Figure 1: A dynamic network environment gives rise to distributional shifts in the data, causing loss of OWD model performance. Continual learning can balance model plasticity and stability, ensuring new information is captured in the model while preserving old knowledge, that is avoiding catastrophic forgetting.
• Figure 2: Multi-generator generative replay architecture. (I) Scholar from the previous task, and inputs and targets in the current task. (II) The procedure of learning of the scholar at the current task, which involves three steps.
• Figure 3: Floor plan of the testbed area illustrating positions and movement patterns (P1 - P6).
• Figure 4: Performance evaluation on CL task sequence Group 1 and Group 2.
• Figure 5: KDE plot of principal components on generated ${\boldsymbol x^\prime}$ by TVAE and VAE and histogram on corresponding ${y^\prime}$.