Code and data

The code below aims to enhance the accessibility and reproducibility of my work. If you use it, please cite the associated papers.

Attention is not all you need
Code reproducing our findings.
code, bibtex, preprint, blogpost

Structural message passing
SMP is a more powerful message passing graph neural network. The respective paper appeared in NeurIPS 2020.
code, bibtex, paper, video, blogpost

Erdos goes neural
The code for the respective paper from NeurIPS 2020.
code, bibtex, paper, video, blogpost

Graph coarsening
Obtain coarse graphs that are spectrally similar to a target graph and reproduce the results from “Graph reduction with spectral and cut guarantees” JMLR 2019.
code, bibtex, paper, blogpost

Attention & convolution
Code reproducing results from “On the relationship between self-attention and convolution” ICLR 2020.
code, bibtex, paper, blogpost

Solve the path extrapolation problem and reproduce results from “Path extrapolation with graph neural networks” IJCAI 2019.
code, bibtex, paper, blogpost

Joint Fourier Transform
The routines for Fourier analysis of graph signals are part of the GSPBOX. The code below reproduces results from “A time-vertex signal processing framework” TSP 2018.
code, bibtex, preprint

Time-vertex stationarity
Model stochastic graph signals that vary in time and reproduce the results from “Stationary time-vertex signal processing” JASP 2019.
code, bibtex, paper

ARMA graph filters
Use these filters to process graph signals while taking into account long-range interactions between nodes, as in “Autoregressive Moving Average Graph Filtering” TSP 2017.
code, bibtex, preprint, blogpost

Independent implementations and applications: ARMA convolution (Pytorch Geometric, Spektral, see also this); SPINNER graph partitioning on GIRAPH (Okapi)

All code is distributed freely.