Introducing Routing Uncertainty in Capsule Networks

Fabio De Sousa Ribeiro, Georgios Leontidis, Stefanos Kollias

Research output: Contribution to conferencePaperpeer-review

111 Downloads (Pure)

Abstract

Rather than performing inefficient local iterative routing between adjacent capsule layers, we propose an alternative global view based on representing the inherent uncertainty in part-object assignment. In our formulation, the local routing iterations are replaced with variational inference of part-object connections in a probabilistic capsule network, leading to a significant speedup without sacrificing performance. In this way, global context is also considered when routing capsules by introducing global latent variables that have direct influence on the objective function, and are updated discriminatively in accordance with the minimum description length (MDL) principle. We focus on enhancing capsule network properties, and perform a thorough evaluation on pose-aware tasks, observing improvements in performance over previous approaches whilst being more computationally efficient.
Original languageEnglish
Pages1-13
Number of pages13
Publication statusPublished - 5 Nov 2020
Event34th Conference on Neural Information Processing Systems (NeurIPS 2020) - Virtual, Vancouver, Canada
Duration: 6 Dec 202012 Dec 2020
https://neurips.cc/

Conference

Conference34th Conference on Neural Information Processing Systems (NeurIPS 2020)
Abbreviated titleNeurIPS
CountryCanada
CityVancouver
Period6/12/2012/12/20
Internet address

Keywords

  • Deep Learning
  • Capsule Networks

Fingerprint Dive into the research topics of 'Introducing Routing Uncertainty in Capsule Networks'. Together they form a unique fingerprint.

Cite this