Graph Neural Networks for Information Extraction with PyTorch

Explore graph neural networks (GNNs) for information extraction in this 28-minute EuroPython 2021 conference talk. Gain insights into GNNs as a generalization of convolutional neural networks (CNNs) used in computer vision. Learn how to conceptualize images as graphs, with nodes representing pixels and edges depicting neighbor relationships. Discover the transition from image graphs to arbitrary graphs, leading to basic GNN architecture. Survey Python implementations and supporting libraries, focusing on PyTorch framework and PyTorch Geometric library. Delve into NLP applications, particularly information extraction from tabular documents. Understand how graphs encode spatial word disposition for training deep learning models with improved accuracy and generalization. Cover topics including the deep learning landscape, GNN use cases, information extraction pipelines, graph representation in Python, and comparisons with convolutional networks.

Intro
Deep learning landscape
Sample GNN use case
A typical information extraction pipeline
What is a graph?
Representing graphs in Python
Graph neural networks
Comparison with convolutional networks
Use case: information extraction from tables
Use case: a possible model architecture
Implementations, literature