Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Liquid Neural Network-based Adaptive Learning vs. Incremental Learning for Link Load Prediction amid Concept Drift due to Network Failures

Omran Ayoub, Davide Andreoletti, Aleksandra Knapińska, Róża Goścień, Piotr Lechowicz, Tiziano Leidi, Silvia Giordano, Cristina Rottondi, Krzysztof Walkowiak

TL;DR

This work tackles concept drift in network Traffic prediction after link failures by comparing liquid neural networks (LNNs) that adapt online to sudden distribution shifts with an incremental-learning baseline that retrains periodically. The authors model traffic on the Euro28 topology using a sine-based traffic generator and simulate failures that induce sharp pattern changes, evaluating RMSE, MAPE, and Time to Convergence. Results show that LNNs excel under drastic drift, providing rapid, retraining-free adaptation, while incremental learning can outperform LNNs under moderate drift when retraining is appropriately scheduled; a hybrid approach that leverages both strengths is suggested. The study offers practical guidance for operators on deploying adaptive forecasting to speed network restoration and improve utilization during failure conditions.

Abstract

Adapting to concept drift is a challenging task in machine learning, which is usually tackled using incremental learning techniques that periodically re-fit a learning model leveraging newly available data. A primary limitation of these techniques is their reliance on substantial amounts of data for retraining. The necessity of acquiring fresh data introduces temporal delays prior to retraining, potentially rendering the models inaccurate if a sudden concept drift occurs in-between two consecutive retrainings. In communication networks, such issue emerges when performing traffic forecasting following a~failure event: post-failure re-routing may induce a drastic shift in distribution and pattern of traffic data, thus requiring a timely model adaptation. In this work, we address this challenge for the problem of traffic forecasting and propose an approach that exploits adaptive learning algorithms, namely, liquid neural networks, which are capable of self-adaptation to abrupt changes in data patterns without requiring any retraining. Through extensive simulations of failure scenarios, we compare the predictive performance of our proposed approach to that of a reference method based on incremental learning. Experimental results show that our proposed approach outperforms incremental learning-based methods in situations where the shifts in traffic patterns are drastic.

Liquid Neural Network-based Adaptive Learning vs. Incremental Learning for Link Load Prediction amid Concept Drift due to Network Failures

TL;DR

This work tackles concept drift in network Traffic prediction after link failures by comparing liquid neural networks (LNNs) that adapt online to sudden distribution shifts with an incremental-learning baseline that retrains periodically. The authors model traffic on the Euro28 topology using a sine-based traffic generator and simulate failures that induce sharp pattern changes, evaluating RMSE, MAPE, and Time to Convergence. Results show that LNNs excel under drastic drift, providing rapid, retraining-free adaptation, while incremental learning can outperform LNNs under moderate drift when retraining is appropriately scheduled; a hybrid approach that leverages both strengths is suggested. The study offers practical guidance for operators on deploying adaptive forecasting to speed network restoration and improve utilization during failure conditions.

Abstract

Adapting to concept drift is a challenging task in machine learning, which is usually tackled using incremental learning techniques that periodically re-fit a learning model leveraging newly available data. A primary limitation of these techniques is their reliance on substantial amounts of data for retraining. The necessity of acquiring fresh data introduces temporal delays prior to retraining, potentially rendering the models inaccurate if a sudden concept drift occurs in-between two consecutive retrainings. In communication networks, such issue emerges when performing traffic forecasting following a~failure event: post-failure re-routing may induce a drastic shift in distribution and pattern of traffic data, thus requiring a timely model adaptation. In this work, we address this challenge for the problem of traffic forecasting and propose an approach that exploits adaptive learning algorithms, namely, liquid neural networks, which are capable of self-adaptation to abrupt changes in data patterns without requiring any retraining. Through extensive simulations of failure scenarios, we compare the predictive performance of our proposed approach to that of a reference method based on incremental learning. Experimental results show that our proposed approach outperforms incremental learning-based methods in situations where the shifts in traffic patterns are drastic.
Paper Structure (17 sections, 1 equation, 3 figures, 1 table)

This paper contains 17 sections, 1 equation, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Addressing Concept Drift
  4. ML-based Traffic Prediction
  5. Adaptive Learning and Incremental Learning for Link Load Prediction
  6. Reference Approach: Incremental Learning
  7. Proposed Approach: LNN-based Online Learning
  8. Traffic and Network Model
  9. Traffic Model
  10. Network Model
  11. Experimental Results
  12. Experimental Settings
  13. Failure Scenarios
  14. Model Parameters and Training
  15. Evaluation Metrics
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Example of a concept drift in traffic patterns and trends on a link due to network failure (at time step 100).
  • Figure 2: Topology of the network and failure scenarios.
  • Figure 3: The rmse and mape achieved by Online LNN, Incremental-20 and Incremental-5 along the 50 time steps after the failure event (i.e., after concept drift) for the highly impacted (subfigures (a) and (b)) and moderately impacted (subfigures (c) and (d)) cases.