News

The preprint introduces a model for random abstract cell complexes and a method to sample 2-dimensional cell complexes. We've updated the preprint after improving the accuracy of the approximate sampling and the presentation in the paper.

We have also highlighted the use case of this approach for (approximately) calculating properties over the cycles on a graph. To this end, the companion python package py-raccoon now exposes intermediate data.

The preprint is available on arXiv and py-raccoon on PyPI.

We have just minted a new PhD! Michael Scholkemper today successfully defended his thesis “On the Structural Analysis of Nodes in Networks”, covering topics ranging from the assignment of roles to nodes to deep learning on graphs.

We wish him all the best in his new position at the DZNE in Bonn.

Congratulations Dr. Scholkemper!

Alexandre René’s paper “Selecting fitted models under epistemic uncertainty using a stochastic process on quantile functions” is now finally out in Nature Communications!

We are increasingly seeing data-driven methods used to fit scientific models, but methods to compare these models often underestimate relevant uncertainties, leading to overconfident comparisons. This paper examines how one can improve the quantification of uncertainties, especially those arise from modelling approximations or errors, and thus make statements about the most likely generative process underlying those data. The theory can be broadly applied to any type generative model, from graphs to neuroscience.

Our paper "Faster Inference of Cell Complexes from Flows via Matrix Factorization" got accepted at EUSIPCO 2025.

In this paper, we consider the problem of inferring 2-cells from signals observed on the edges of a graph, s.t. the signals can be represented as a sparse combination of gradient and curl flows (see also our previous paper). We show matrix factorization to lead to an efficient heuristic for inferring said 2-cells.

We are happy to announce the HOOC workshop on higher-order networks. Attendance is free! Go to https://conf.netsci.rwth-aachen.de for more information and registration.

Network analysis has revolutionized our understanding of complex systems, and graph-based methods have emerged as powerful tools to process signals on non-Euclidean domains via graph signal processing and graph neural networks. However, graphs are ill-equipped to encode multi-way and higher-order relations – features that are essential to understanding many systems such as group-dynamics in social systems, multi-gene interactions in genetic data, or multi-way drug interactions.

Accordingly, there is a need for new analytical methods to address the challenges of higher-order data, and a growing body of work in this direction. With this workshop, we especially want to explore current challenges which arise when bridging theory and data. This means on the one hand discussing what higher-order methods have already been used with real-world data, and on the other hand, what challenges currently prevent modelling systems with higher-order interactions.

The workshop will take place from the 11th to the 13th of August 2025. We look forward to welcoming you in Aachen, Germany!

Our paper "HLSAD: Hodge Laplacian-based Simplicial Anomaly Detection" has beeen accepted at KDD 2025, taking place in Toronto, Canada from August 2-7 this year.

HLSAD is a novel event and change-point detection algorithm for time-evolving simplicial complexes. It leverages the Hodge Laplacian to capture higher-order topological features and detect anomalies in the dynamic data by analyzing the evolution of the Hodge Laplacian spectrum. We show the effectiveness of out approach for both graph-lifting and inherently higher-order scenarios.

The preprint is available on arXiv and source code on GitLab.

We have just minted a new PhD! Felix Stamm today successfully defended his thesis “Models and Algorithms for Systematic Network Randomization”, on methods for randomizing graphs while preserving different aspects of their local structure. These were applied to problems both of creating null graph models and anonymizing graph information.

We wish him all the best in his future endeavours.

Congratulations Dr. Stamm!

Our paper "Learning From Simplicial Data Based on Random Walks and 1D Convolutions" has been accepted at ICLR 2024.

In this paper, we propose a learning algorithm on topological domains based on random walks, which are processed by 1D convolutional neural networks. We show that this approach outperforms existing methods such as SCNN and MPSN on several datasets.

The paper is available on OpenReview and source code on GitLab.