Group Equivariant Subsampling

Abstract

Subsampling is used in convolutional neural networks (CNNs) in the form of pooling or strided convolutions, to reduce the spatial dimensions of feature maps and to allow the receptive fields to grow exponentially with depth. However, it is known that such subsampling operations are not translation equivariant, unlike convolutions that are translation equivariant. Here, we first introduce translation equivariant subsampling/upsampling layers that can be used to construct exact translation equivariant CNNs. We then generalise these layers beyond translations to general groups, thus proposing group equivariant subsampling/upsampling. We use these layers to construct group equivariant autoencoders (GAEs) that allow us to learn low-dimensional equivariant representations. We empirically verify on images that the representations are indeed equivariant to input translations and rotations, and thus generalise well to unseen positions and orientations. We further use GAEs in models that learn object-centric representations on multi-object datasets, and show improved data efficiency and decomposition compared to non-equivariant baselines.

References in corpus (20)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Rotation-invariant convolutional neural networks for galaxy morphology prediction
• Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
• Disentangling by Factorising
• Group Equivariant Convolutional Networks
• Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
• Making Convolutional Networks Shift-Invariant Again
• InfoVAE: Information Maximizing Variational Autoencoders
• Object-Centric Learning with Slot Attention
• MONet: Unsupervised Scene Decomposition and Representation
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Locally Scale-Invariant Convolutional Neural Networks
• General $E(2)$-Equivariant Steerable CNNs
• Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
• Spatial Broadcast Decoder: A Simple Architecture for Learning Disentangled Representations in VAEs
• Intertwiners between Induced Representations (with Applications to the Theory of Equivariant Neural Networks)
• Attentive Group Equivariant Convolutional Networks
• Co-Attentive Equivariant Neural Networks: Focusing Equivariance On Transformations Co-Occurring In Data
• Spin-Weighted Spherical CNNs