Optimizing Diffusion Rate and Label Reliability in a Graph-Based Semi-supervised Classifier

Afonso, Bruno Klaus de Aquino; Berton, Lilian

doi:10.1007/978-3-030-91702-9_34

Cited by 3 publications

(1 citation statement)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, we should always refer to the baseline prediction as a feature, not a label. The reason is twofold: irst, the actual label in this problem is the path delay obtained in the simulation conducted by the organizers; secondly, the prediction will be available for all future data, which avoids some over itting issues when using known labels to optimize graph models with gradient descent [12].…”

Section: Qt-routenet: Model Overviewmentioning

confidence: 99%

QT-Routenet: Improved GNN generalization to larger 5G networks by fine-tuning predictions from queueing theory

Afonso¹,

Berton²

2022

ITU J-FET

View full text Add to dashboard Cite

In order to promote the use of machine learning in 5G, the International Telecommunication Union (ITU) proposed in 2021 the second edition of the ITU AI/ML in 5G challenge, with over 1600 participants from 82 countries. This work details the second place solution overall, which is also the winning solution of the Graph Neural Networking Challenge 2021. We tackle the problem of generalization when applying a model to a 5G network that may have longer paths and larger link capacities than the ones observed in training. To achieve this, we propose to first extract robust features related to Queueing Theory (QT), and then fine-tune the analytical baseline prediction using a modification of the Routenet Graph Neural Network (GNN) model. The proposed solution generalizes much better than simply using Routenet, and manages to reduce the analytical baseline's 10.42 mean absolute percent error to 1.45 (1.27 with an ensemble). This suggests that making small changes to an approximate model that is known to be robust can be an effective way to improve accuracy without compromising generalization.

show abstract

Section: Qt-routenet: Model Overviewmentioning

confidence: 99%