The recent wave of impressive results obtained in fields as varied as computer vision, natural language processing, bioinformatics and many more can be attributed to the advances in training and designing neural networks. A neural network works as a universal function approximator, so that it can use training data to learn complex input-output mappings. This chapter presents spectral approaches to the definition of graph-convolutional layers, drawing from literature on the graph Fourier transform (GFT). It focuses on the spatial definitions of graph convolution, which have emerged as a more flexible alternative, providing superior experimental performance. The spectral approach to graph convolution is desirable as it is mathematically principled, relying on the spectral domain induced by the GFT. The graph convolution network has been applied in numerous applications and it is one of the most well-known graph convolutional neural networks. Computational complexity is a major issue, especially for adaptive approaches deriving weights as functions of features.

Graph Neural Networks / Fracastoro, Giulia; Valsesia, Diego - In: Graph Spectral Image ProcessingSTAMPA. - [s.l] : Wiley, 2021. - ISBN 9781789450286. - pp. 63-72 [10.1002/9781119850830.ch3]

Graph Neural Networks

Giulia Fracastoro;Diego Valsesia
2021

Abstract

The recent wave of impressive results obtained in fields as varied as computer vision, natural language processing, bioinformatics and many more can be attributed to the advances in training and designing neural networks. A neural network works as a universal function approximator, so that it can use training data to learn complex input-output mappings. This chapter presents spectral approaches to the definition of graph-convolutional layers, drawing from literature on the graph Fourier transform (GFT). It focuses on the spatial definitions of graph convolution, which have emerged as a more flexible alternative, providing superior experimental performance. The spectral approach to graph convolution is desirable as it is mathematically principled, relying on the spectral domain induced by the GFT. The graph convolution network has been applied in numerous applications and it is one of the most well-known graph convolutional neural networks. Computational complexity is a major issue, especially for adaptive approaches deriving weights as functions of features.
2021
9781789450286
9781119850830
Graph Spectral Image Processing
File in questo prodotto:
File Dimensione Formato  
gsip_ch3.pdf

non disponibili

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Non Pubblico - Accesso privato/ristretto
Dimensione 215.07 kB
Formato Adobe PDF
215.07 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2977426