Using and Abusing Equivariance
In this paper we show how Group Equivariant Convolutional Neural Networks use subsampling to learn to break equivariance to their symmetries. We focus on 2D rotations and reflections and investigate the impact of broken equivariance on network performance. We show that a change in the input dimension of a network as small as a single pixel can be enough for commonly used architectures to become approximately equivariant, rather than exactly. We investigate the impact of networks not being exactly equivariant and find that approximately equivariant networks generalise significantly worse to unseen symmetries compared to their exactly equivariant counterparts. However, when the symmetries in the training data are not identical to the symmetries of the network, we find that approximately equivariant networks are able to relax their own equivariant constraints, causing them to match or outperform exactly equivariant networks on common benchmark datasets.
Code (0)
등록된 구현이 없습니다.
Methods 이 논문이 사용한 방법론
Similar Papers 제목 키워드 기반
TrollMeta@DravidianLangTech-EACL2021: Meme classification using deep learning
Memes act as a medium to carry one’s feelings, cultural ideas, or practices by means of symbols, imitations, or simply images. Whenever social media is involved, hurting the feelings of others and abusing others are alwa…
ClassificationDeep LearningMeme ClassificationAbusing Images and Sounds for Indirect Instruction Injection in Multi-Modal LLMs
We demonstrate how images and sounds can be used for indirect prompt and instruction injection in multi-modal LLMs. An attacker generates an adversarial perturbation corresponding to the prompt and blends it into an imag…
What Affects Learned Equivariance in Deep Image Recognition Models?
Equivariance w.r.t. geometric transformations in neural networks improves data efficiency, parameter efficiency and robustness to out-of-domain perspective shifts. When equivariance is not designed into a neural network,…
Data AugmentationInductive BiasTranslationEquivariance-aware Architectural Optimization of Neural Networks
Incorporating equivariance to symmetry groups as a constraint during neural network training can improve performance and generalization for tasks exhibiting those symmetries, but such symmetries are often not perfectly n…
Neural Architecture SearchRelaxing Equivariance Constraints with Non-stationary Continuous Filters
Equivariances provide useful inductive biases in neural network modeling, with the translation equivariance of convolutional neural networks being a canonical example. Equivariances can be embedded in architectures throu…
image-classificationImage Classification