MICCAI 2019: A Tutorial on Hypergraph Neural Network Toolbox

Abstract

We organize a tutorial about “Hypergraph Learning: Methods, Tools and Applications in Medical Image Analysis”. I give a talk about “A Tutorial on Hypergraph Neural Network Toolbox”.

Date
Oct 13, 2019 8:00 AM — 11:20 AM
Location
Shenzhen, China

THU-DeepHypergraph(THU-DH) is a python toolbox for hypergraph-based deep learning, which is built upon pytorch. Edge in hypergraph named hyperedge can link more than two nodes, which allows hyperedge to express more than pair-wise relation(like: entity-attribute relation, group relation, hierarchical relation and so on.). Thus, hypergraph owns more powerful model ability than common graph. It consists of sparse hypergraph construction, fusion, convolution operations, convenient util functions for medical image(MRI, Pathology, etc.), 3D(point cloud, view-based graph, etc.) and other hypergraph applications(to be continue…). Hypergrpah inductive learning and hypergraph transductive learning examples is also included in this toolbox. What’s more, we write several examples that deploy hypergraph in different tasks like: Classification, Segmentation and Regression.

Related