Introducing Routing Uncertainty in Capsule Networks
- Submitting institution
-
University of Aberdeen
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 177346561
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- 34th Conference on Neural Information Processing Systems (NeurIPS 2020)
- First page
- 1
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- November
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
-
- Citation count
- -
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- The proposed approach is a step change towards the wider adoption of capsule networks for real-life computer vision tasks, as it achieves a significant training speedup, whilst improving performance and introducing the novel concept of routing uncertainty. This output is highly significant because it demonstrates via a rigorous process and experimentation that the whole concept around routing for training capsule networks needs rethinking. It then goes on to propose a novel solution that addresses these issues, whilst retaining the superiority and advantage over convolutional neural networks in out-of-distribution generalisation, affine transformation robustness, and generalisation to novel viewpoints.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -
