A Simple Strategy to Provable Invariance via Orbit Mapping
Not on the map. Why?
Abstract (arXiv)
Many applications require robustness, or ideally invariance, of neural networks to certain transformations of input data. Most commonly, this requirement is addressed by training data augmentation, using adversarial training, or defining network architectures that include the desired invariance by design. In this work, we propose a method to make network architectures provably invariant with respect to group actions by choosing one element from a (possibly continuous) orbit based on a fixed criterion. In a nutshell, we intend to 'undo' any possible transformation before feeding the data into the actual network. Further, we empirically analyze the properties of different approaches which incorporate invariance via training or architecture, and demonstrate the advantages of our method in terms of robustness and computational efficiency. In particular, we investigate the robustness with respect to rotations of images (which can hold up to discretization artifacts) as well as the provable orientation and scaling invariance of 3D point cloud classification.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 51 of its 83 references that are in the map. First posted on arXiv 24 Sep 2022.
Builds on
51 of 83 references.
References · 51
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2014Adam: A Method for Stochastic Optimizationcited by 37,240
- 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
- 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,248
- 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,401
- 2018Unsupervised Representation Learning by Predicting Image Rotationscited by 1,591
- 2015Spatial Transformer Networkscited by 1,429
- 2016Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzlescited by 1,382
- 2016Group Equivariant Convolutional Networkscited by 1,165
- 2021E(n) Equivariant Graph Neural Networkscited by 965
- 2018Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point cloudscited by 746
- 2020SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networkscited by 607
38 more
- 2019Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimationcited by 495
- 2016Geometric deep learning on graphs and manifolds using mixture model CNNscited by 468
- 2019General E(2)-Equivariant Steerable CNNscited by 446
- 2018Rotation Equivariant CNNs for Digital Pathologycited by 422
- 2021High-Performance Large-Scale Image Recognition Without Normalizationcited by 360
- 20183D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Datacited by 354
- 2016Harmonic Networks: Deep Translation and Rotation Equivariancecited by 295
- 2017Learning SO(3) Equivariant Representations with Spherical CNNscited by 248
- 2014Understanding image representations by measuring their equivariance and equivalencecited by 243
- 2017Equivariance Through Parameter-Sharingcited by 141
- 2018PPF-FoldNet: Unsupervised Learning of Rotation Invariant 3D Local Descriptorscited by 141
- 2017Deep Functional Maps: Structured Prediction for Dense Shape Correspondencecited by 134
- 20183D Point Capsule Networkscited by 106
- 2016Rotation equivariant vector field networkscited by 99
- 2017Polar Transformer Networkscited by 81
- 2017Exploring the Landscape of Spatial Robustnesscited by 81
- 2019AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations rather than Datacited by 77
- 2020Deep Shells: Unsupervised Shape Correspondence with Optimal Transportcited by 74
- 2016TI-POOLING: transformation-invariant pooling for feature learning in Convolutional Neural Networkscited by 67
- 2018Learning Invariances using the Marginal Likelihoodcited by 51
- 2020Certified Defense to Image Transformations via Randomized Smoothingcited by 51
- 2020CaSPR: Learning Canonical Spatiotemporal Point Cloud Representationscited by 51
- 2014Deep Roto-Translation Scattering for Object Classificationcited by 50
- 2018A Graph-CNN for 3D Point Cloud Classificationcited by 49
- 2020On Isometry Robustness of Deep 3D Point Cloud Models under Adversarial Attackscited by 46
- 2019Equivariant Transformer Networkscited by 35
- 2022ConDor: Self-Supervised Canonicalization of 3D Pose for Partial Shapescited by 31
- 2020Learning to Orient Surfaces by Self-supervised Spherical CNNscited by 26
- 2016Learning rotation invariant convolutional filters for texture classificationcited by 24
- 2016Warped Convolutions: Efficient Invariance to Spatial Transformationscited by 24
- 2020Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysiscited by 22
- 2019Invariance-inducing regularization using worst-case transformations suffices to boost accuracy and spatial robustnesscited by 18
- 2019Endowing Deep 3D Models with Rotation Invariance Based on Principal Component Analysiscited by 17
- 2020Isometric Transformation Invariant and Equivariant Graph Convolutional Networkscited by 17
- 2021Learning Rotation-Invariant Representations of Point Clouds Using Aligned Edge Convolutional Neural Networkscited by 12
- 2022ART-Point: Improving Rotation Robustness of Point Cloud Classifiers via Adversarial Rotationcited by 12
- 2020Geometric Adversarial Attacks and Defenses on 3D Point Cloudscited by 11
- 2019OperatorNet: Recovering 3D Shapes From Difference Operatorscited by 5
50 most cited of 51 All 51 (JSON)
Built on by
Citing papers per year: 2022: 1; 2023: 1.
2 papers.