Adoption in Open-Source Software
XGI · HAT · HIF
Hypergraph Centrality
Our node and edge nonlinear eigenvector centrality for hypergraphs (Tudisco & Higham, Communications Physics 2021) is implemented in several widely used hypergraph libraries:
- XGI (CompleX Group Interactions), one of the main Python libraries for higher-order networks, ships it as
node_edge_centrality. - HAT – Hypergraph Analysis Toolbox (Python and MATLAB), developed by researchers at the University of Michigan, Harvard Medical School, Raytheon Technologies Research Center and iReprogram (PLOS Computational Biology 2023), implements it as
nonlinear_eigenvector_centrality. - The cross-library Hypergraph Interchange Format (HIF) paper, written by the developers of HyperNetX (PNNL), XGI, Hypergraphx, HAT and SimpleHypergraphs.jl, uses nonlinear eigenvector centrality as its showcase example.
Hypergraphx
Core–Periphery Detection in Hypergraphs
Hypergraphx (HGX)’s core–periphery function is documented as an “Implementation of the core-periphery model described in” our paper (Tudisco & Higham, SIAM J. Mathematics of Data Science 2023).
DHG-Bench
Hypergraph Neural Networks
HyperND, our nonlinear feature diffusion on hypergraphs (ICML 2022), is one of the models in DHG-Bench (ICLR 2026), a benchmark of hypergraph neural networks. It is also a standard baseline in hypergraph learning; for example, ED-HNN calls it “the SOTA hypergraph diffusion algorithm”.
National Labs and Industry
National lab · US Department of Energy
Dynamical Low-Rank Training at Oak Ridge National Laboratory
Dynamical Low-Rank Training (DLRT) (Low-rank lottery tickets, NeurIPS 2022) and its follow-ups have become a research line at Oak Ridge National Laboratory (US Department of Energy), led by our co-author Steffen Schotthöfer:
- A NeurIPS 2025 oral paper (arXiv) on low-rank compression with adversarial robustness. It builds on DLRT, our robust low-rank training method (NeurIPS 2023) and our Tucker extension (NeurIPS 2024), and is funded by ORNL’s Laboratory Directed Research and Development programme.
- Federated dynamical low-rank training (FeDLRT).
- Compressing geospatial vision transformers for edge devices with DLRT.
Independent groups also extend DLRT, including the Karlsruhe Institute of Technology (arXiv 2401.08505) and a NeurIPS 2026 paper on stabilised DLRT.
Industry
Featured by G-Research
G-Research, a headline sponsor of NeurIPS 2022, picked our Low-rank lottery tickets paper for its NeurIPS 2022 paper reviews. In the review, one of its quantitative researchers walks through how the method trains networks directly in low-rank form.
NVIDIA · Sony AI
GeoLoRA
GeoLoRA (ICLR 2025), our geometry-aware method for parameter-efficient fine-tuning, is used as one of the two key adaptive-rank baselines by researchers at NVIDIA and the University of Michigan in MARS (2026). It is also cited as the earlier three-factor orthogonal formulation in Sony AI’s StelLA (NeurIPS 2025).
Industry
Industry Research Building on Our Low-Rank Work
Our low-rank training papers are cited by research from:
- Amazon and EleutherAI (ReLoRA)
- Tencent AI Lab (LORO, ICLR 2025)
- Mitsubishi Electric Research Labs (LatentLLM)
- Intel (the FETTA training accelerator)
Heliophysics
Solar Physics: NASA and IBM
Surya, the heliophysics foundation model from NASA and IBM Research, cites our Solaris foundation model of the Sun as prior work on forecasting solar activity.
Finance · BBVA Research
Systemic Risk in Banking
A BBVA Research working paper (July 2026) on systemic risk in Argentina’s credit network models firms that borrow from several banks as hyperedges. It implements our three hypergraph centrality models as benchmark measures and describes them in a dedicated appendix.
Quantum computing
Quantum Optimisation with Leonardo
With Leonardo S.p.A.’s quantum computing lab, we studied drone-delivery scheduling on a D-Wave quantum annealer (Quantum Machine Intelligence 2026) and on Pasqal’s neutral-atom quantum processor. Our earlier QUBO formulation of core–periphery detection (KDD 2022) is cited by D-Wave in Milestones on the Quantum Utility Highway (ACM Transactions on Quantum Computing, 2023).
Competition
First Place at ACM ICAIF 2025
My PhD student Kaicheng (Kevin) Zhang and Denis Zorba won first place at the ACM ICAIF 2025 Cryptocurrency Forecasting Competition (about 130 participants, more than 740 models). The winning model was our Stuart–Landau oscillatory graph neural network.
Applications Across the Sciences
Protein structure
A Nature Communications study of the “protein universe” computes our hypergraph centrality (max variant) to score each residue’s structural influence. The scores flag structurally important regions and likely sites of damaging mutations (Nature Communications 2025).
Biology
Researchers at the University of Queensland extended our hypergraph centrality, reusing our code, to rank proteins in a yeast protein-complex network. The rankings predict which proteins are essential and extend the “centrality–lethality rule” to hypergraphs (PLOS ONE 2024).
Neuroscience
Our nonlinear spectral core–periphery method (SIMODS 2019) was used to extract features from dynamic brain connectivity, classifying sleep-deprivation states with 92% accuracy (Brain Sciences 2019).
Political science
A study of civic life in a French village computes our centralities for every resident. The nonlinear variant best tracks political participation among association members (PLOS ONE 2025).
Finance
A new centrality for S&P 100 asset-correlation networks builds explicitly on our hypergraph centrality (Economies 2024). See also the BBVA study above.
Social media
Aalto University researchers used our fast Chung–Lu generator (Networks 2020) to benchmark polarisation measures, which they then applied to 203 Twitter networks (ACM CSCW 2022).
In Landmark Reviews and Books
Reference reviews citing our hypergraph work
- What are higher-order networks? (SIAM Review 2023)
- The structure and dynamics of networks with higher order interactions (Physics Reports 2023)
- Epidemic spreading on higher-order networks (Physics Reports 2024)
- Higher-order interactions shape collective human behaviour (Nature Human Behaviour 2025)
- Contagion dynamics on higher-order networks (Nature Reviews Physics 2024)
- Collective dynamics on higher-order networks (Nature Reviews Physics 2026)
- A Survey on Hypergraph Mining (ACM Computing Surveys 2025)
Books and handbooks citing our work
- Qi, Chen & Chen, Tensor Eigenvalues and Their Applications (Springer, 2018)
- Couillet & Liao, Random Matrix Methods for Machine Learning (Cambridge University Press, 2022)
- Elsevier’s Handbook of Numerical Analysis (2022)
- Springer’s Handbook of Mathematical Models and Algorithms in Computer Vision and Imaging
- Holme & Saramäki (eds.), Temporal Network Theory (Springer, 2nd ed. 2023)
Our spectral core–periphery method is included in A clarified typology of core–periphery structure in networks (Science Advances 2021). Our nonlinear spectral theory is covered in the 2025 SIAM Review survey on The Gross–Pitaevskii equation and eigenvector nonlinearities (Henning & Jarlebring), which cites four of our papers.
On the deep-learning side, DLRT is covered in A Survey of Lottery Ticket Hypothesis and in LLM Inference Unveiled (Tencent AI Lab, UC Berkeley and others).
Media Coverage
December 2025
Physics World
Covered our study of how remote working changed scientific collaboration and impact (article; paper).
February 2026
Quantum Zeitgeist
Covered our drone-delivery work with Leonardo on a neutral-atom quantum computer (article).
September–October 2024
Retraction Watch & infoDOCKET
Retraction Watch (Sep 2024) and Library Journal’s infoDOCKET (Oct 2024) featured our Nature Human Behaviour paper What we should learn from pandemic publishing. That paper has an Altmetric score in the top 7% of research outputs of its age.
On social media
- Scientific Reports highlighted our paper Collaboration and topic switches in science twice (Jan 2024, Mar 2024).
- The SIAM Activity Group on Dynamical Systems featured our review Nonlinear spectral graph theory (Apr 2025, about 4,700 views) and our Science Advances paper on the effective order of hypergraph dynamics (May 2024).
- Satoshi Matsuoka, Director of the RIKEN Center for Computational Science, shared our Low-rank lottery tickets announcement (Sep 2022).
- The XGI developers announced their implementation of our hypergraph centrality (Jul 2022).
- The ICAIF 2025 win was announced by the competition organisers and the School of Mathematics.
Awards and Recognition
SIAM SIGEST Award (2023)
Our SIMAX paper with Antoine Gautier and Matthias Hein was selected for SIGEST and republished in expanded form as Nonlinear Perron–Frobenius Theorems for Nonnegative Tensors (SIAM Review 2023).
Most-cited SIMAX papers
Two of our papers appear in SIAM News’s list of the 20 most-cited SIMAX papers since 2018, listed first and third.
Plenary talks
At the Householder Symposium XXI (2022), IC2S2 2023 (one of 16 plenaries selected from more than 900 submissions) and the SOCN Graduate School study day (2025).
Editor’s Choice
Medical & Biological Engineering & Computing (March 2026), for our Graph-Convolutional-β-VAE for synthetic aortic aneurysm generation.
Fellowships
- Turing Fellow, Alan Turing Institute (2024)
- Bayes Innovation Fellow (2024)
- GAIL Fellow, Edinburgh Generative AI Lab (2024)
- Marie Skłodowska-Curie Fellow (MAGNET, 2017–2019)
Service
- Associate Editor, SIAM Review
- Managing Editor, ETNA
- Area Chair, ICLR, NeurIPS and ICML
- Science Communication Coordinator, COST Action mSPACE (CA24122)
From Research to Practice
Co-founded 2023 · Chief Scientific Officer
Miniml.AI
I co-founded Miniml.AI in 2023 and serve as its Chief Scientific Officer. Miniml brings our work on efficient, trustworthy AI into enterprise and healthcare settings, and its research arm, Miniml Labs, publishes alongside the group.
Bayes Innovation Fellowship
Commercialisation
As a Bayes Innovation Fellow, I worked on developing our model-compression methods, which reach up to 95% compression on large computer-vision tasks, into a commercial platform. I’ve also lead a University of Edinburgh Big Ideas Accelerator project on faithful and trustworthy generative AI for healthcare.
People
Former PhD students from the group have gone on to:
University of Oxford
Emanuele Zangrando
Postdoc on the mathematical foundations of AI, Mathematical Institute, University of Oxford.
University of Edinburgh
Sara Tarquini
Research Associate, Quantum Software Lab, University of Edinburgh.
Brookhaven National Laboratory
Harris Abdul Majid
Brookhaven National Laboratory (US Department of Energy).
Uniklinik RWTH Aachen
Dayana Savostianova
Scientist in applied AI for digital pathology, Laboratory of Nephropathology, Uniklinik RWTH Aachen.
Northeastern University
Sara Venturini
MOBS Lab, Network Science Institute, Northeastern University, after a postdoc at MIT Senseable City Lab.
RWTH Aachen University
Piero Deidda
Researcher in explainable ML & AI, RWTH Aachen University.
Cite this page
@misc{tudisco2026impact,
author = {Francesco Tudisco},
title = {Real-World Impact of Our Research},
year = {2026},
howpublished = {\url{https://ftudisco.gitlab.io/impact/}}
}