Leveraging Graph Neural Networks to Parametrize IoT Protocols

In the ever-growing realm of IoT, IoT protocols are vital, fueling a wide range of applications. Fine-tuning these protocols by means of selecting a good set of parameters is very important to ensure high performance. Typically, optimization of protocol parameters based on data collected using testbeds is preferred, as it reflects real-world uncertainties. However, the exhaustive testing of all possible parameter combinations is a time-intensive task that is exacerbated when a protocol should be tested with several network layouts and topologies. In such cases, exhaustive testing is unfeasible, and other techniques should be investigated. Our goal is to employ graph neural networks (GNNs) and leverage their innate ability to utilize the graph-like data structures that wireless networks naturally form. GNNs are particularly suited,as they excel at unraveling the complex relationships and spatial dependencies inherent among the nodes. The strategy involves harnessing data derived from the parameterization of various network layouts for a given protocol to train a GNN model. This model shall parse this data and learn from it, enabling the prediction of the most effective protocol parameters for new, previously unseen network layouts. By adopting this method, we anticipate a significant reduction in the time required to optimize IoT protocol configurations for new layouts.

Download as PDF

Student Target Groups:

  • Students of ICE/Telematics;
  • Students of Computer Science;
  • Students of Electrical Engineering.

Thesis Type:

  • Master Thesis / Master Project

Goal and Tasks:

  • Getting familiar with state-of-the-art IoT protocols and measure their performance using real-world testbeds;
  • Investigate and identify key features of wireless networks that in uence protocol performance, focusing on parameters that are crucial for efficient operation;
  • Explore effective methodologies for encoding network features, aiming to maximize their representational effciency for model training;
  • Utilize graph neural networks (GNNs) to model wireless network configurations, leveraging their ability to process complex network structures;
  • Validate the GNN model's predictions against actual testbed results to assess accuracy and practical applicability.

Recommended Prior Knowledge:

  • Basic knowledge of machine learning;
  • Solid skills in Python;
  • Basic knowledge of networked systems.


  • a.s.a.p.