Capsule Networks overcome some shortcomings of convolutional neural networks organizing neurons into groups of capsules. Capsule layers are dynamically connected by means of an iterative routing mechanism, which models the connection strengths between capsules from different layers. However, whether routing improves the network performance is still object of debate. This work tackles this issue via Routing Annealing (RA), where the number of routing iterations is annealed at training time. This proposal gives some insights on the effectiveness of the routing for Capsule Networks. Our experiments on different datasets and architectures show that RA yields better performance over a reference setup where the number of routing iterations is fixed (even in the limit case with no routing), especially for architectures with fewer parameters.

Capsule Networks with Routing Annealing

Renzulli R.;Tartaglione E.;Fiandrotti A.;Grangetto M.
2021-01-01

Abstract

Capsule Networks overcome some shortcomings of convolutional neural networks organizing neurons into groups of capsules. Capsule layers are dynamically connected by means of an iterative routing mechanism, which models the connection strengths between capsules from different layers. However, whether routing improves the network performance is still object of debate. This work tackles this issue via Routing Annealing (RA), where the number of routing iterations is annealed at training time. This proposal gives some insights on the effectiveness of the routing for Capsule Networks. Our experiments on different datasets and architectures show that RA yields better performance over a reference setup where the number of routing iterations is fixed (even in the limit case with no routing), especially for architectures with fewer parameters.
2021
30th International Conference on Artificial Neural Networks, ICANN 2021
virtual
2021
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Springer Science and Business Media Deutschland GmbH
12891
529
540
978-3-030-86361-6
978-3-030-86362-3
Annealing; Capsule networks; Routing
Renzulli R.; Tartaglione E.; Fiandrotti A.; Grangetto M.
File in questo prodotto:
File Dimensione Formato  
ICANN21_capsule.pdf

Accesso aperto

Tipo di file: POSTPRINT (VERSIONE FINALE DELL’AUTORE)
Dimensione 687.57 kB
Formato Adobe PDF
687.57 kB Adobe PDF Visualizza/Apri

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/2318/1844236
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact