A Receptor Skeleton for Capsule Neural Networks

Jintai Chen, Hongyun Yu, Chengde Qian, Danny Z Chen, Jian Wu
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:1781-1790, 2021.

Abstract

In previous Capsule Neural Networks (CapsNets), routing algorithms often performed clustering processes to assemble the child capsules’ representations into parent capsules. Such routing algorithms were typically implemented with iterative processes and incurred high computing complexity. This paper presents a new capsule structure, which contains a set of optimizable receptors and a transmitter is devised on the capsule’s representation. Specifically, child capsules’ representations are sent to the parent capsules whose receptors match well the transmitters of the child capsules’ representations, avoiding applying computationally complex routing algorithms. To ensure the receptors in a CapsNet work cooperatively, we build a skeleton to organize the receptors in different capsule layers in a CapsNet. The receptor skeleton assigns a share-out objective for each receptor, making the CapsNet perform as a hierarchical agglomerative clustering process. Comprehensive experiments verify that our approach facilitates efficient clustering processes, and CapsNets with our approach significantly outperform CapsNets with previous routing algorithms on image classification, affine transformation generalization, overlapped object recognition, and representation semantic decoupling.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-chen21x, title = {A Receptor Skeleton for Capsule Neural Networks}, author = {Chen, Jintai and Yu, Hongyun and Qian, Chengde and Chen, Danny Z and Wu, Jian}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {1781--1790}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/chen21x/chen21x.pdf}, url = {https://proceedings.mlr.press/v139/chen21x.html}, abstract = {In previous Capsule Neural Networks (CapsNets), routing algorithms often performed clustering processes to assemble the child capsules’ representations into parent capsules. Such routing algorithms were typically implemented with iterative processes and incurred high computing complexity. This paper presents a new capsule structure, which contains a set of optimizable receptors and a transmitter is devised on the capsule’s representation. Specifically, child capsules’ representations are sent to the parent capsules whose receptors match well the transmitters of the child capsules’ representations, avoiding applying computationally complex routing algorithms. To ensure the receptors in a CapsNet work cooperatively, we build a skeleton to organize the receptors in different capsule layers in a CapsNet. The receptor skeleton assigns a share-out objective for each receptor, making the CapsNet perform as a hierarchical agglomerative clustering process. Comprehensive experiments verify that our approach facilitates efficient clustering processes, and CapsNets with our approach significantly outperform CapsNets with previous routing algorithms on image classification, affine transformation generalization, overlapped object recognition, and representation semantic decoupling.} }
Endnote
%0 Conference Paper %T A Receptor Skeleton for Capsule Neural Networks %A Jintai Chen %A Hongyun Yu %A Chengde Qian %A Danny Z Chen %A Jian Wu %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-chen21x %I PMLR %P 1781--1790 %U https://proceedings.mlr.press/v139/chen21x.html %V 139 %X In previous Capsule Neural Networks (CapsNets), routing algorithms often performed clustering processes to assemble the child capsules’ representations into parent capsules. Such routing algorithms were typically implemented with iterative processes and incurred high computing complexity. This paper presents a new capsule structure, which contains a set of optimizable receptors and a transmitter is devised on the capsule’s representation. Specifically, child capsules’ representations are sent to the parent capsules whose receptors match well the transmitters of the child capsules’ representations, avoiding applying computationally complex routing algorithms. To ensure the receptors in a CapsNet work cooperatively, we build a skeleton to organize the receptors in different capsule layers in a CapsNet. The receptor skeleton assigns a share-out objective for each receptor, making the CapsNet perform as a hierarchical agglomerative clustering process. Comprehensive experiments verify that our approach facilitates efficient clustering processes, and CapsNets with our approach significantly outperform CapsNets with previous routing algorithms on image classification, affine transformation generalization, overlapped object recognition, and representation semantic decoupling.
APA
Chen, J., Yu, H., Qian, C., Chen, D.Z. & Wu, J.. (2021). A Receptor Skeleton for Capsule Neural Networks. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:1781-1790 Available from https://proceedings.mlr.press/v139/chen21x.html.

Related Material