Ideal Machine Intelligence
Home
People
Research
Teaching
Publications
Erik J. Bekkers
🌙 Dark Mode
Teaching
/ GEDL
Group Equivariant
Deep Learning
A graduate-level course on the mathematical foundations of geometric deep learning.
Module 1: Regular Group Convolutional Neural Networks
1.1
Introduction & Motivation
1.2
Group Theory: Groups, Actions, and Representations
1.3
Regular Group Convolutions
1.4
SE(2) NN Example: Histopathology
1.5
A Brief History of G-CNNs
1.6
Group Theory: Transitive Actions & Quotients
1.7
Group Convolutions are All You Need
Module 2: Steerable Group Convolutional Neural Networks
2.1
Steerable Kernels & Basis Functions
2.2
Regular G-Convs with Steerable Kernels
2.3
Group Theory: Irreps & Fourier Transform
2.4
Group Theory: Induced Representations & Fields
2.5
Steerable Group Convolutions
2.6
Activation Functions for Steerable G-CNNs
2.7
Harmonic Networks from Regular G-Convs
Module 3: Equivariant Graph Neural Networks
3.1
Motivation for SE(3) Equivariant GNNs
3.2
Equivariant Message Passing
3.3
Tensor Products as Conditional Linear Layers
3.4
Group Theory: SO(3) Irreps & Clebsch-Gordan
3.5
Literature: 3D Steerable GNNs
3.6
Literature: Regular Equivariant GNNs
3.7
Literature: Gauge Equivariant GNNs