logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-09172019-163550


Tipo di tesi
Tesi di laurea magistrale
Autore
LANDOLFI, FRANCESCO
URN
etd-09172019-163550
Titolo
K-plex Pooling for Graph Deep Neural Networks
Dipartimento
INFORMATICA
Corso di studi
INFORMATICA
Relatori
relatore Prof. Bacciu, Davide
relatore Prof. Grossi, Roberto
relatore Prof. Conte, Alessio
relatore Prof. Marino, Andrea
Parole chiave
  • geometric deep learning
  • graph neural network
  • k-plex
  • pooling
  • pseudoclique
Data inizio appello
04/10/2019
Consultabilità
Non consultabile
Data di rilascio
04/10/2089
Riassunto
This thesis proposes a novel pooling technique for Graph Neural Networks based on k-plexes, i.e., pseudo-cliques where each node can miss up to k links to the other nodes within the same subgraph. This pooling method, named KPlexPool, is completely non-parametrical, and generates a coarsen graph exploiting only its topological information. This model was tested on graph classification tasks using standard benchmark datasets from literature, and its performance was compared with the ones obtained from other parametrical pooling methods, showing state-of-the-art results.
File