#71 - ZAK JOST (Graph Neural Networks + Geometric DL) [UNPLUGGED]
Machine Learning Street Talk (MLST)
00:00
Challenges in Graph Neural Networks
This chapter explores the complexities of applying convolutional techniques in graph structures through graph neural networks (GNNs). It addresses issues like varying neighbor counts, message passing using adjacency matrices, and the trade-offs in expressive power amidst computational challenges. Additionally, it discusses solutions to problems such as over-smoothing and over-squashing while emphasizing the significance of localized information in processing node features.
Transcript
Play full episode
