Research
My name is Pascal Mettes and I am a tenured Assistant Professor at the University of Amsterdam. The mission of me and my team is to advance the field of hyperbolic deep learning. Currently, deep learning is centred around Euclidean geometry. Euclidean geometry however has critical blindspots, ones which cannot be undone by adding more data and making bigger networks. Among the most important issues with modern deep learning is dealing with hierarchies and this has a fundamental reason: hierarchies are hyperbolic in nature due to their exponential growth. We develop the theory and algorithms to perform deep learning in hyperbolic geometry, the natural geometry of hierarchies. Below, I highlight our key research domains.
Vision-language models should be hyperbolic
Vision-language models play a central role in computer vision. The optimization of current vision-language models however contain two crucial mistakes. These mistakes can only be solved through hyperbolic embeddings. Today's vision-language models assume that vision and language are symmetrical and that vision-language alignment is flat. Language is more general than vision however (a picture is worth a thousand words) and vision-language alignment is deeply hierarchical. Hyperbolic vision-language models are the future, see our latest research on this below:
ICLR 2025 "Compositional Entailment Learning for Hyperbolic Vision-Language Models"
(oral) Paper
Avik Pal, Max van Spengler, Guido Maria D'Amely di Melendugno, Alessandro Flaborea, Fabio Galasso, Pascal Mettes
CVPR 2025 "Hyperbolic Safety-Aware Vision-Language Models"
(spotlight) Paper
Tobia Poppi, Tejaswi Kasarla, Pascal Mettes, Lorenzo Baraldi, Rita Cucchiara
TMLR 2024 "Intriguing Properties of Hyperbolic Embeddings in Vision-Language Models"
Paper
Sarah Ibrahimi, Mina Ghadimi Atigh, Nanne Van Noord, Pascal Mettes, Marcel Worring
Hierarchical deep learning with hyperbolic embeddings
Hierarchies exhibit an exponential growth in nodes as a function of depth. Since Euclidean space only grows polynomially with its radius, there is an inherent mismatch between both. Luckily, hyperbolic space also exhibits exponential growth. If you want to do deep learning with hierarchical knowledge, or embed hierarchies in a continuous space, hyperbolic geometry is the way to go. Hierarchical learning is key towards avoiding large errors and generalizing to new classes, as shown in our research highlighted below:
ICML 2025 "Low-distortion and GPU-compatible Tree Embeddings in Hyperbolic Space"
Paper
Max van Spengler, Pascal Mettes
ESWC 2025 "Designing Hierarchies for Optimal Hyperbolic Embedding"
(best paper nomination) Paper
Melika Ayoughi, Max van Spengler, Pascal Mettes, Paul Groth
IJCV 2025 "SimZSL: Zero-Shot Learning Beyond a Pre-defined Semantic Embedding Space"
Paper
Mina Ghadimi Atigh, Stephanie Nargang, Martin Keller-Ressel, Pascal Mettes
CVPR 2022 "Hyperbolic Image Segmentation"
Paper
Mina Ghadimi Atigh, Julian Schoep, Erman Acar, Nanne van Noord, Pascal Mettes
NeurIPS 2021 "Hyperbolic busemann learning with ideal prototypes"
Paper
Mina Ghadimi Atigh, Martin Keller-Ressel, Pascal Mettes
CVPR 2020 "Searching for Actions on the Hyperbole"
Paper
Teng Long, Pascal Mettes, Heng Tao Shen, Cees Snoek
Robust deep learning in hyperbolic space
Current deep networks are not robust. They have trouble dealing with out-of-distribution data, are sensitive to adversarial attacks, are difficult for planning, and can make arbitrarily bad mistakes. In our research, we have found that hyperbolic geometry has the potential to make deep learning far more robust, with out-of-distribution detection as highlight. See below papers on this topic:
IROS 2024 "Hyp2Nav: Hyperbolic Planning and Curiosity for Crowd Navigation"
(oral) Paper
Alessandro Flaborea, Guido Maria D'Amely di Melendugno, Pascal Mettes, Fabio Galasso
TMLR 2024 "Hyperbolic Random Forests"
Paper
Lars Doorenbos, Pablo Márquez-Neila, Raphael Sznitman, Pascal Mettes
ECCVw 2024 "Adversarial Attacks on Hyperbolic Networks"
Paper
Max van Spengler, Jan Zahálka, Pascal Mettes
ICCV 2023 "Poincaré ResNet"
Paper
Max van Spengler, Erwin Berkhout, Pascal Mettes
LOD 2022 "Hyperbolic Graph Codebooks"
(oral) Paper
Pascal Mettes
Resources and reading materials
In our journey towards hyperbolic deep learning, we have compiled a survey, a software library, and hours of video material with context, background, and deeper information on this exciting new research field. Below are recommended materials if you want to get started in the field:
IJCV 2024 "Hyperbolic Deep Learning in Computer Vision: A Survey"
Paper
Pascal Mettes, Mina Ghadimi Atigh, Martin Keller-Ressel, Jeffrey Gu, Serena Yeung
MM 2023 "HypLL: The Hyperbolic Learning Library"
Paper Code
Max van Spengler, Philipp Wirth, Pascal Mettes
ECCV 2022 "Hyperbolic Representation Learning for Computer Vision"
Youtube
Pascal Mettes, Mina Ghadimi Atigh, Martin Keller-Ressel, Jeffrey Gu, Serena Yeung
Recent talks on hyperbolic learning (2024-2025)
KeynoteSydney (AU) Non-Euclidean Foundation Models and Geometric Learning workshop, the Web Conference
KeynotePrague (CZ) 49th Pattern Recognition and Computer Vision Colloquium
KeynoteAberdeen (UK) BMVA Summer School
Planned
Invited talkBristol (UK) University of Bristol
Planned
Invited talk Vienna (AT) ELLIS workshop Institute of Science and technology Austria
Planned
Invited talkVienna (AT) University of Vienna
Planned
Invited talkLeuven (BE) KU Leuven
Invited talkDelft (NL) Delft University of Technology
KeynoteOxford (UK) Oxford Machine Learning Summer School
KeynoteTokyo (JP) 1st Workshop on Human-Centered Vision and Media Technologies
KeynoteTokyo (JP) AIST Computer Vision Challenges Workshop
Invited talkAmsterdam (NL) VU Amsterdam
Invited talkHeidelberg (DE) University of Heidelberg
Students
PhD students on hyperbolic learning
Max van Spengler Co-supervisor: prof. Erwin Berkout 2022-now
Tejaswi Kasarla Co-supervisor: prof. Rita Cucchiara 2022-now
Melika Ayoughi Co-supervisor: prof. Paul Groth 2019-now
Mina Ghadimi Atigh 2019-now
HAVA Lab and AI4Film PhD students
Anne Zonneveld Co-supervisor: dr. Iris Groen 2024-now
Gowreesh Mago Co-supervisor: dr. Stevan Rudinac 2024-now
Swasti Mishra Co-supervisor: prof. Erwin Berkhout 2024-now
Carlo Bretti Co-supervisor: dr. Nanne van Noord 2023-now
Visitors
Stefano D'Arrigo Sapienza University of Rome 2025
Teng Long University of Amsterdam 2024
Alessandro Flaborea Sapienza University of Rome 2024
Guido D'Amely Sapienza University of Rome 2024
Academic activities, grants, and awards
Teaching
MSc Artificial IntelligenceCoordinator Deep Learning 1 2024-now
MSc Artificial IntelligenceGuest lecturer Foundation Models 2025
BSc/MSc Information StudiesChair Education Committee 2019-2025
MSc Information StudiesCoordinator Applied Machine Learning 2018-2024
MSc Artificial IntelligenceCoordinator Thesis AI 2019-2022
MSc Artificial IntelligenceMember Entry committee MSc AI 2019-2021
MSc Artificial IntelligenceGuest lecturer Computer Vision I and II 2018-2021
Academic service
Program Chair International Conference on Multimedia Retrieval 2026
General chair Netherlands Conference on Computer Vision 2022
Organizer 2nd workshop Beyond Euclidean: Hyperbolic and Hyperspherical Learning for Computer Vision ICCV 2025
Organizer 1nd workshop Beyond Euclidean: Hyperbolic and Hyperspherical Learning for Computer Vision ECCV 2024
Organizer Hyperbolic Deep Learning for Computer Vision workshop CVPR 2023
Organizer Hyperbolic Representation Learning for Computer Vision workshop ECCV 2022
Area Chair CVPR 2023-now, NeurIPS 2023-now, ECCV 2024-now, ICML 2025-now, BMVC 2020-now
Web Chair ACM Multimedia 2016
Grants
PI ELLIs PhD Award co-PI: prof. Rita Cucchiara
PI NWO ClickNL
PI Google Perception Academic Funding with: Thomas Mensink
PI Data Science Centre PhD Grant co-PI: prof. Erwin Berkhout
PI Informatics Institute PhD Collaboration Grant Fellow PI: prof. Paul Groth
co-PI HAVA Lab PI: prof. Cees Snoek
co-PI NWO ClickNL PI: dr. Nanne van Noord
co-PI ACTA AiO Competition PI: prof. Bruno Loos, Erwin Berkhout
Awards
Best paper nominationESWC25 for "Designing Hierarchies for Optimal Hyperbolic Embedding"
FinalistMM 2023 Best Open-Source Software Competition (for HypLL)
AwardCVPR21, ICLR21, ECCV20, ICML20, NeurIPS21, ICMR17, ICCV17, ICMR15 Reviewer award
AwardMM 2016 Best Doctoral Student Award
AwardTRECVID 2015 Winner Multimedia Event Detection Benchmark