🐣 Welcome to the lab, Olivia. We are glad you are joining us!
Welcome to the AIDOS Lab
Data has shape.
We build the tools that find it.
We are a research group at the University of Fribourg, working at the intersection of geometry, topology, and machine learning. Grounded in mathematics, we favor simplicity, elegance, and interpretability over mere performance.
If you are a student at the University of Fribourg and are interested in writing a bachelor’s or master’s thesis with us, please drop us a line.
What We Do
Our core focus is geometrical and topological machine learning, i.e., developing methods that make use of principles from geometry and topology to learn robust, expressive representations. We work with concepts like Euler Characteristic Transforms, persistent homology, discrete curvature, and metric space magnitude to analyze point clouds, graphs, and manifolds.
We see ourselves as toolsmiths, caring about theory and practice alike.
Our research is graciously supported by the Canton of Fribourg and the Swiss State Secretariat for Education, Research and Innovation SERI, which funds our ERC Starting Grant “HOLES: Higher-Order Learning of Essential Structures with Geometry and Topology” under the transitional measures for the Horizon package 2021–2027.
Why We Do It
“AIDOS” carries two meanings. The first one describes our work, i.e., Artificial Intelligence for Data-Oriented Science. The second one originates from an ancient Greek word:
αἰδώς: a sense of awe, reverence, or humility when facing something greater than oneself.
This awe keeps us honest about the many things we do not (yet) know, and we strongly believe that this is the right disposition for doing science.
News
🚀 Jeremy defended his PhD thesis “Understanding Data Representations using Geometry and Topology” with distinction (magna cum
laude). What a feat, Jeremy! It was a pleasure working with you. All
the best for your future endeavors and may we find many opportunities
to reconnect in the future.
🚀 Ernst defended his PhD thesis “Towards Deep Learning with Euler
Characteristic Transforms” with distinction (magna cum laude). Congratulations,
Ernst—it was great working with you. All the best and may our paths
cross again.
🐣 Juan joins the lab. A warm welcome, Juan—we are very happy to
have you!
🎉 Our curvature-based dimensionality reduction method Provable cluster-preserving visualizations with curvature-based stochastic neighbor embeddings is now officially out in PNAS.
🚀 Katharina received a CRCHUM Postdoctoral Fellowship, enabling her to
move forward with new exciting projects. We are very excited for this
next step and wish you all the best!
🎉 Our work on scale discontinuities of GNNs, entitled Graph Neural Networks Are Not Continuous Across Graph Resolutions,
was accepted at ICML. Congratulations, Christian!
🎉 Our work on EmbedOR: Provable Cluster-Preserving Visualizations with Curvature-Based Stochastic Neighbor Embeddings was
accepted at PNAS! A wonderful collaboration with Tristan Saidi,
Abigail Hickok, and Andrew Blumberg.
🎉 Our work on TOAST: Transformer Optimization using Adaptive and
Simple Transformations
was accepted at TMLR. Congratulations to our academic visitor and
collaborator Irene!
Rubén defended his dissertation “Topology-Enhanced Deep Learning” with a cum laude certification.
He has joined Axiom to focus on theorem
proving and AI for Mathematics. Congratulations and all the best—it
was an honor working with you!
Julius defended his dissertation “Robust Topological Representation
Learning” with magna cum laude and is the first PhD student of the
lab to graduate. Congratulations and all the best for your future
career, Julius! It was a pleasure working with you!
🎉 Our work on LEAP: Local ECT-Based Learnable Positional Encodings
for Graphs was accepted
to ICLR 2026.
🎉 Our work on Molecular Machine Learning Using Euler Characteristic
Transforms,
was published. This was spearheaded by our academic visitor
Victor—congratulations!
👏 Our academic visitor Giacomo (Parolin) defended his MSc thesis
“Differentiable Euler Characteristic Transform for Molecular
Prediction” with the highest possible exam grade “cum laude.”
Congratulations!
🎉 Jeremy’s paper Strategies to accelerate US coal power phase-out
using contextual retirement
vulnerabilities
appeared in Nature Energy. Congratulations!
🎉 Our research on geometry-aware edge
pooling, LLM analysis using
intrinsic dimensionality, and
point cloud generation was
accepted at NeurIPS 2025. Our small lab has now been present in all
of the three major machine learning conferences since its foundation
in 2022. May Fortuna continue smiling upon us! 🍀
🎉 Johannes’s paper Stable and Accurate Orbital-Free Density
Functional Theory Powered by Machine
Learning appeared in
the Journal of the American Chemical Society. Congratulations!
🎉 Bastian’s article Topology Meets Machine Learning: An Introduction Using the Euler Characteristic Transform
in the Notices of the AMS is out.
If you are wondering what this is all about, here are brief summary
posts on X
and BlueSky.
🎙 Bastian was interviewed on the DataSkeptic podcast
to discuss our work on No Metric to Rule Them All: Toward Principled Evaluations of Graph-Learning Datasets.
🎉 Our works on local variants of the Euler Characteristic Transform
and on a principled analysis of graph-learning data sets
have been accepted at ICML 2025!
🎉 Our works on molecule generation
and on new data sets for topological deep learning
have been accepted at ICLR 2025!
Bastian will give three talks at
JMM, the Joint Mathematics
Meetings, one on “Diss-lECT: Dissecting Data with local Euler Characteristic
Transforms” (related to a recent preprint of
ours), the second one on “Two
Households, Both Alike In Dignity: Geometry and Topology in Machine
Learning,” and the final one on “Good Gradients and How To Find Them:
Towards Multi-Scale Representation Learning.”
Find these (and more!) talks at Bastian’s website.