HLA @ AMLD EPFL 2025

Modern deep learning in computer vision is dominated by networks designed to operate on Euclidean manifolds. But is Euclidean geometry truly the best fit for these models, or just a convenient choice? Recent advances in machine learning and computer vision suggest that hyperbolic geometry presents a powerful alternative. This geometry offers enhanced capabilities for embedding complex structures such as hierarchies, graphs, text, images, and videos.

In light of these developments, our tutorial aims to introduce hyperbolic geometry as a promising tool for computer vision, appealing to both researchers and practitioners. Whether you're new to the concept or looking to deepen your understanding, this tutorial offers both theoretical and practical insights.

What to Expect

At the conference, we’ll provide an accessible introduction to hyperbolic geometry, designed especially for non-mathematicians. Our goal is to focus on intuition and high-level concepts, ensuring that the topic is approachable and engaging for everyone.

We will also cover the latest advancements in the field, exploring the current use of hyperbolic geometry in computer vision from both supervised and unsupervised perspectives. By the end of the session, we’ll highlight open research challenges and the future potential of hyperbolic geometry for visual understanding.

Practical Insights and Hands-On Experience

Unlike typical theoretical tutorials, we go beyond the foundations. Our tutorial website will feature interactive, notebook-style code snippets based on foundational works in hyperbolic geometry. These resources are designed to help you gain hands-on experience, lowering the barrier to entry and allowing you to actively explore this exciting new research area.

Whether you're just starting your journey or are already familiar with deep learning, this tutorial is your guide to understanding and utilizing hyperbolic geometry in computer vision.