ICCV 2025 Workshop

Beyond Euclidean: Hyperbolic and Hyperspherical Learning for Computer Vision

2nd Edition

19-20 October, Honolulu Hawaii

Call for papers, submission portal is now open! See here for more details.

Bringing together researchers to uncover the principles of Non-euclidean representations.

Within deep learning, Euclidean geometry is the default basis for deep neural networks, yet the naive assumption that such a topology is optimal for all data types and tasks does not necessarily hold. There exists a growing body of evidence to suggest that data and the representations that we aim to learn can be better captured through learning in corresponding geometries when exhibiting non-Euclidean structures. The interest in non-Euclidean deep learning has grown dramatically in recent years, with advancing methodologies, libraries, and applications. Beyond Euclidean will be the first workshop solely devoted to deep learning in hyperbolic and hyperspherical spaces.

Hyperbolic and hyperspherical learning are quickly gaining traction in computer vision and deep learning. Hyperbolic and hyperspherical learning approaches bring new perspectives and have the ability to address open issues in conventional deep learning. Hyperbolic geometry can be viewed as the natural geometry of hierarchies (Mettes et. al. 23) and has recently shown to be a highly effective approach for problems ranging from few-shot recognition (Khrulkov et al. 20) to out-of-distribution detection (Guo et.al. 22, Flaborea et.al. 23), self-supervised learning and much more. Similarly, hyperspherical learning is omnipresent in today's contrastive learning, which powered by cosine similarities, and has deep impact in all tasks from self-supervision (Durrant et.al. 22) to long-tailed classification (Kasarla et.al. 22) and learning with limited samples (Trosten et. al. 23).

The Beyond Euclidean workshop bring together computer vision researchers with a shared interest in exploring non-Euclidean geometry. The workshop invites researchers to submit their latest work, fostering engaging discussions through invited and contributed talks. We aim to overcome the limitations of Euclidean representations and unlock new possibilities.

Call for papers: join us to challenge conventional perspectives and shape the future of computer vision!

Invited Speakers

Carlo Vittorio Cannistraci

Carlo is a theoretical engineer and computational innovator. He is a Chair Professor in the Tsinghua Laboratory of Brain and Intelligence (THBI) and adjunct professor in the Department of Computer Science and the School of Biomedical Engineering at Tsinghua University. He directs the Center for Complex Network Intelligence (CCNI) in THBI, which aims to create pioneering algorithms at the interface between information science, physics of complex systems, complex networks and machine intelligence, with a focus on brain/life-inspired computing for efficient artificial intelligence and big data analysis. These computational methods are often applied to precision biomedicine, neuroscience, social and economic science.

Mehrtash Harandi

Mehrtash is an Associate Professor in the Department of Electrical and Computer Systems Engineering (ECSE), Monash University, Australia. His research focuses on machine learning, computer vision, and optimization, with particular interests in learning with limited supervision, continual and lifelong learning, optimization over structured spaces (e.g., Riemannian geometry), and responsible AI (e.g., machine unlearning). His talk will focus on Poincaré Kernels for Hyperbolic Representations.

Aditya sinha

Aditya is a Research Scientist at Netflix Research in the Foundation Models and Inference Research group. His research focuses on the practice and theory of training and post-training, alignment, and evaluation of LLMs. Prior to this, Aditya held fellow and engineer roles at Microsoft and Google, respectively. His recent research into structured representations and use of Hyperbolic embeddings, notably within industry research, will contribute to the workshop extensively.