Fast prototyping of GNNs for communication networks
IGNNITION is the ideal framework for beginners in neural network programming
Developed with passion by networking enthusiasts for scientists and engineers of the field
Build a custom GNN adapted to your particular network scenario
Design and run your own GNN model in a matter of few hours with three simple steps:
- Define a GNN architecture with an intuitive YAML interface
- Adapt your dataset
- Execute the training with just 3 lines of code
IGNNITION will produce an optimized implementation of your GNN without writing a single line of TensorFlow
Why IGNNITION
High-level abstraction
IGNNITION introduces MSMP graphs, a novel abstraction that represents visually GNN models. This allows you to define your own model in a YAML file without considering the underlying mathematical formulation (e.g. tensor-wise operations).
No TensorFlow code
IGNNITION targets users with no experience on neural network programming. Define easily your GNN model via a declarative language in YAML. Then, the model can be trained/evaluated with just 3 lines of code. Other popular GNN frameworks (e.g., DGL, PyTorch Geometric) require to deal with complex tensor-based programming.
High flexibility of design
IGNNITION provides a modular design interface that supports all the popular GNN architectures and enables to combine individual components of them (e.g., GCN, attention, multi-stage message passing, heterogeneous entities). Check out our library with popular GNNs already implemented!
Easy debugging
Graph Neural Networks are not black boxes! IGNNITION produces interactive visual representations that let you dive into your GNN design. It also offers an advanced error-checking system that identifies bugs and shows how to fix them.
Easy integration
IGNNITION provides an interface to easily feed your model with datasets from different sources and in various formats. Also, it offers a user-friendly API that facilitates the integration of GNNs with other solutions (e.g., Deep Reinforcement Learning).
High performance
IGNNITION is implemented on top of TensorFlow. It has been tested over many GNN architectures and the results show that it is as efficient as native Tensorflow implementations coded by experts.Join our community
IGNNITION is an open source project that loves your comments and contributions. Please join our community
This project has received funding from the European Union’s Horizon 2020 research and innovation programme within the framework of the NGI-POINTER Project funded under grant agreement No 871528.
Citing
Plain text:
David Pujol-Perich, José Suárez-Varela, Miquel Ferriol, Shihan Xiao, Bo Wu, Albert Cabellos-Aparicio, and Pere Barlet-Ros. 2021. IGNNITION: Bridging the Gap between Graph Neural Networks and NetworkingSystems. IEEE Network 35, 6 (2021), 171–177.
BibTeX:
@article{pujol2021ignnition,
title={IGNNITION: Bridging the Gap between Graph Neural Networks and Networking Systems},
author={Pujol-Perich, David and Su{\'a}rez-Varela, Jos{\'e} and Ferriol, Miquel and Xiao, Shihan and Wu, Bo and Cabellos-Aparicio, Albert and Barlet-Ros, Pere},
journal={IEEE Network},
volume={35},
number={6},
pages={171--177},
year={2021},
publisher={IEEE},
doi={10.1109/MNET.001.2100266}
}