This repository contains the code for:
L. Fesser, B. Kiani, and M. Weber. Unitary Convolutions for Message-passing and Positional Encodings on Directed Graphs. International Conference on Machine Learning (ICML), 2026.
Dune supports experiments on both directed and undirected graph datasets.
For experiments on datasets with directed edges, see the directed_mpnns subfolder and its README.
For experiments on undirected graphs that study the role of edge features, see the undirected_experiments subfolder and its README.
The experiments have been tested on NVIDIA A100 and H100 GPUs with CUDA 12.9.
If you use this codebase, please cite:
@inproceedings{fesser2026unitary,
title={Unitary Convolutions for Message-passing and Positional Encodings on Directed Graphs},
author={Fesser, Lukas and Kiani, Bobak and Weber, Melanie},
booktitle={International Conference on Machine Learning},
year={2026}
}