Learning Stable Group Invariant Representations with Convolutional Networks

Abstract

Transformation groups, such as translations or rotations, effectively express part of the variability observed in many recognition problems. The group structure enables the construction of invariant signal representations with appealing mathematical properties, where convolutions, together with pooling operators, bring stability to additive and geometric perturbations of the input. Whereas physical transformation groups are ubiquitous in image and audio applications, they do not account for all the variability of complex signal classes. We show that the invariance properties built by deep convolutional networks can be cast as a form of stable group invariance. The network wiring architecture determines the invariance group, while the trainable filter coefficients characterize the group action. We give explanatory examples which illustrate how the network architecture controls the resulting invariance group. We also explore the principle by which additional convolutional layers induce a group factorization enabling more abstract, powerful invariant representations.

4 pages

Cited by in corpus (16)

• Understanding Deep Convolutional Networks
• Robust Large Margin Deep Neural Networks
• Deep Neural Networks with Random Gaussian Weights: A Universal Classification Strategy?
• Scattering Networks for Hybrid Representation Learning
• Transformation Properties of Learned Visual Representations
• Generalization Error of Invariant Classifiers
• Algebraic Neural Networks: Stability to Deformations
• Warped Convolutions: Efficient Invariance to Spatial Transformations
• Scaling the Scattering Transform: Deep Hybrid Networks
• On Deep Representation Learning from Noisy Web Images
• Building a Regular Decision Boundary with Deep Networks
• Convolutional Filtering and Neural Networks with Non Commutative Algebras
• On the effect of pooling on the geometry of representations
• Emergence of Selective Invariance in Hierarchical Feed Forward Networks
• Quantised Transforming Auto-Encoders: Achieving Equivariance to Arbitrary Transformations in Deep Networks
• How ConvNets model Non-linear Transformations