Geometric Deep Learning: a Temperature Based Analysis of Graph Neural Networks

TL;DR

This paper models graph neural networks as thermodynamic systems, analyzing the temperature across layers to gain insights into their behavior and potential applications.

Contribution

It introduces a novel thermodynamic perspective to analyze GCN and GAT models using the concept of temperature across layers.

Findings

Temperature varies across layers, revealing insights into model dynamics.

Potential applications include improved understanding and optimization of GNNs.

Provides a new framework for analyzing neural network behavior.

Abstract

We examine a Geometric Deep Learning model as a thermodynamic system treating the weights as non-quantum and non-relativistic particles. We employ the notion of temperature previously defined in [7] and study it in the various layers for GCN and GAT models. Potential future applications of our findings are discussed.