Compositionally Equivariant Representation Learning
Not on the map. Why?
Abstract (arXiv)
Deep learning models often need sufficient supervision (i.e. labelled data) in order to be trained effectively. By contrast, humans can swiftly learn to identify important anatomy in medical images like MRI and CT scans, with minimal guidance. This recognition capability easily generalises to new images from different medical facilities and to new tasks in different settings. This rapid and generalisable learning ability is largely due to the compositional structure of image patterns in the human brain, which are not well represented in current medical models. In this paper, we study the utilisation of compositionality in learning more interpretable and generalisable representations for medical image segmentation. Overall, we propose that the underlying generative factors that are used to generate the medical images satisfy compositional equivariance property, where each factor is compositional (e.g. corresponds to the structures in human anatomy) and also equivariant to the task. Hence, a good representation that approximates well the ground truth factor has to be compositionally equivariant. By modelling the compositional representations with learnable von-Mises-Fisher (vMF) kernels, we explore how different design and learning biases can be used to enforce the representations to be more compositionally equivariant under un-, weakly-, and semi-supervised settings. Extensive results show that our methods achieve the best performance over several strong baselines on the task of semi-supervised domain-generalised medical image segmentation. Code will be made publicly available upon acceptance at https://github.com/vios-s.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 40 of its 58 references that are in the map. First posted on arXiv 13 Jun 2023.
Builds on
40 of 58 references.
References · 40
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2014Adam: A Method for Stochastic Optimizationcited by 36,960
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
- 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
- 2016V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentationcited by 2,284
- 2020FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidencecited by 2,080
- 2016Temporal Ensembling for Semi-Supervised Learningcited by 1,054
- 2020Concept Bottleneck Modelscited by 1,022
- 2017Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learningcited by 918
- 2022Compositional Visual Generation with Composable Diffusion Modelscited by 619
- 2021Semi-Supervised Semantic Segmentation with Cross Pseudo Supervisioncited by 462
- 2019Domain Generalization by Solving Jigsaw Puzzlescited by 449
28 more
- 2019Domain Generalization via Model-Agnostic Learning of Semantic Featurescited by 405
- 2018Towards a Definition of Disentangled Representationscited by 294
- 2021FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Spacecited by 284
- 2020Weakly-Supervised Disentanglement Without Compromisescited by 237
- 2017Emergence of Invariance and Disentanglement in Deep Representationscited by 213
- 2021FSDR: Frequency Space Domain Randomization for Domain Generalizationcited by 163
- 2020Shape-aware Meta-learning for Generalizing Prostate MRI Segmentation to Unseen Domainscited by 124
- 2020Domain Generalization for Medical Imaging Classification with Linear-Dependency Regularizationcited by 122
- 2020Uncertainty-aware multi-view co-training for semi-supervised medical image segmentation and domain adaptationcited by 81
- 2021Learning to Compose Visual Relationscited by 72
- 2019Disentangled Representation Learning in Cardiac Image Analysiscited by 69
- 2022Enhancing Pseudo Label Quality for Semi-Supervised Domain-Generalized Medical Image Segmentationcited by 51
- 2018Learning Compositional Representations for Few-Shot Recognitioncited by 50
- 2020Compositional Convolutional Neural Networks: A Deep Architecture with Innate Robustness to Partial Occlusioncited by 47
- 2021Self-Supervised Learning Disentangled Group Representation as Featurecited by 47
- 2021Learning Disentangled Representations in the Imaging Domaincited by 46
- 2021Semi-supervised Meta-learning with Disentanglement for Domain-generalised Medical Image Segmentationcited by 36
- 2023Learning Visual Representations via Language-Guided Samplingcited by 32
- 2021Compositional Transformers for Scene Generationcited by 24
- 2021Interactive Disentanglement: Learning Concepts by Interacting with their Prototype Representationscited by 24
- 2016Emergence of Compositional Representations in Restricted Boltzmann Machinescited by 23
- 2020Robust Instance Segmentation through Reasoning about Multi-Object Occlusioncited by 23
- 2017Teaching Compositionality to CNNscited by 18
- 2021Domain Composition and Attention for Unseen-Domain Generalizable Medical Image Segmentationcited by 16
- 2020Disentangled Representations for Domain-generalized Cardiac Segmentationcited by 14
- 2020Measuring the Biases and Effectiveness of Content-Style Disentanglementcited by 14
- 2022vMFNet: Compositionality Meets Domain-generalised Segmentationcited by 11
- 2021Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentationcited by 9
Built on by
Citing papers per year: 2023: 1; 2024: 1.
2 papers.