Building Deep, Equivariant Capsule Networks

Capsule networks are constrained by the parameter-expensive nature of their layers, and the general lack of provable equivariance guarantees. We present a variation of capsule networks that aims to remedy this. We identify that learning all pair-wise part-whole relationships between capsules of successive layers is inefficient. Further, we also realise that the choice of prediction networks and the routing mechanism are both key to equivariance. Based on these, we propose an alternative framework for capsule networks that learns to projectively encode the manifold of pose-variations, termed the space-of-variation (SOV), for every capsule-type of each layer. This is done using a trainable, equivariant function defined over a grid of group-transformations. Thus, the prediction-phase of routing involves projection into the SOV of a deeper capsule using the corresponding function. As a specific instantiation of this idea, and also in order to reap the benefits of increased parameter-sharing, we use type-homogeneous group-equivariant convolutions of shallower capsules in this phase. We also introduce an equivariant routing mechanism based on degree-centrality. We show that this particular instance of our general model is equivariant, and hence preserves the compositional representation of an input under transformations. We conduct several experiments on standard object-classification datasets that showcase the increased transformation-robustness, as well as general performance, of our model to several capsule baselines.

[1]  Geoffrey E. Hinton,et al.  Matrix capsules with EM routing , 2018, ICLR.

[2]  Pascal Libuschewski,et al.  Group Equivariant Capsule Networks , 2018, NeurIPS.

[3]  Andrea Vedaldi,et al.  Understanding Image Representations by Measuring Their Equivariance and Equivalence , 2014, International Journal of Computer Vision.

[4]  Konstantinos N. Plataniotis,et al.  Brain Tumor Type Classification via Capsule Networks , 2018, 2018 25th IEEE International Conference on Image Processing (ICIP).

[5]  Mohammad Taha Bahadori,et al.  Spectral Capsule Networks , 2018 .

[6]  Geoffrey E. Hinton Some Demonstrations of the Effects of Structural Descriptions in Mental Imagery , 1979, Cogn. Sci..

[7]  Geoffrey E. Hinton,et al.  Dynamic Routing Between Capsules , 2017, NIPS.

[8]  Max Welling,et al.  Steerable CNNs , 2016, ICLR.

[9]  Alex Krizhevsky,et al.  Learning Multiple Layers of Features from Tiny Images , 2009 .

[10]  Marc Van Droogenbroeck,et al.  HitNet: a neural network with capsules embedded in a Hit-or-Miss layer, extended with hybrid data augmentation and ghost capsules , 2018, ArXiv.

[11]  Qiang Liu,et al.  An Optimization View on Dynamic Routing Between Capsules , 2018, ICLR.

[12]  Alex Lamb,et al.  Deep Learning for Classical Japanese Literature , 2018, ArXiv.

[13]  Ulas Bagci,et al.  Capsules for Object Segmentation , 2018, ArXiv.

[14]  Gideon Kowadlo,et al.  Sparse Unsupervised Capsules Generalize Better , 2018, ArXiv.

[15]  Xiaowei Zhou,et al.  Polar Transformer Networks , 2017, ICLR.

[16]  Chen Xu,et al.  MS-CapsNet: A Novel Multi-Scale Capsule Network , 2018, IEEE Signal Processing Letters.

[17]  Hyun Seo,et al.  Attention Routing Between Capsules , 2019, 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW).

[18]  Ranga Rodrigo,et al.  DeepCaps: Going Deeper With Capsule Networks , 2019, 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[19]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[20]  Abhinav Dhall,et al.  Dense and Diverse Capsule Networks: Making the Capsules Learn Better , 2018, ArXiv.

[21]  Edson Borin,et al.  The Multi-Lane Capsule Network , 2019, IEEE Signal Processing Letters.

[22]  Gabriel J. Brostow,et al.  CubeNet: Equivariance to 3D Rotation and Translation , 2018, ECCV.

[23]  Chenchen Liu,et al.  How convolutional neural networks see the world - A survey of convolutional neural network visualization methods , 2018, Math. Found. Comput..

[24]  Maurice Weiler,et al.  A General Theory of Equivariant CNNs on Homogeneous Spaces , 2018, NeurIPS.

[25]  Lorenzo Torresani,et al.  STAR-Caps: Capsule Networks with Straight-Through Attentive Routing , 2019, NeurIPS.

[26]  Taewon Jeong,et al.  Ladder Capsule Network , 2019, ICML.

[27]  Kumara Kahatapitiya,et al.  Context-Aware Multipath Networks , 2019, ArXiv.

[28]  Max Welling,et al.  Group Equivariant Convolutional Networks , 2016, ICML.

[29]  Andrew Y. Ng,et al.  Reading Digits in Natural Images with Unsupervised Feature Learning , 2011 .

[30]  Roland Vollgraf,et al.  Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms , 2017, ArXiv.

[31]  Peter Bailis,et al.  Equivariant Transformer Networks , 2019, ICML.

[32]  Geoffrey E. Hinton,et al.  Transforming Auto-Encoders , 2011, ICANN.