Geometry for AI
组织者
演讲者
Daniel Platt
时间
2025年11月27日 17:30 至 18:30
地点
A3-4-301
线上
Zoom 293 812 9202
(BIMSA)
摘要
In this talk I will give an overview over applications of geometry to AI. I plan to mention the following topics:
- Invariant/equivariant machine learning,
- Convolutional Neural Networks (CNNs) on manifolds,
- Graph Neural Networks (GNNs),
- Graph curvature and graph rewiring,
- Geometry of embeddings,
- Optimal transport and Wasserstein distance,
- Riemannian optimisation,
- Learning of symmetries,
- Persistent homology,
- Kernels on manifolds.
Each of these are big research areas, so I will say quite little about each of them. I plan to give one definition and then introduce the main paper(s) in this area as well as current research directions, to empower interested listeners to learn more about the topic and maybe start a project in this area themselves.
- Invariant/equivariant machine learning,
- Convolutional Neural Networks (CNNs) on manifolds,
- Graph Neural Networks (GNNs),
- Graph curvature and graph rewiring,
- Geometry of embeddings,
- Optimal transport and Wasserstein distance,
- Riemannian optimisation,
- Learning of symmetries,
- Persistent homology,
- Kernels on manifolds.
Each of these are big research areas, so I will say quite little about each of them. I plan to give one definition and then introduce the main paper(s) in this area as well as current research directions, to empower interested listeners to learn more about the topic and maybe start a project in this area themselves.