Leveraging Systematic Knowledge of 2D Transformations
Not on the map. Why?
Abstract (arXiv)
The existing deep learning models suffer from out-of-distribution (o.o.d.) performance drop in computer vision tasks. In comparison, humans have a remarkable ability to interpret images, even if the scenes in the images are rare, thanks to the systematicity of acquired knowledge. This work focuses on 1) the acquisition of systematic knowledge of 2D transformations, and 2) architectural components that can leverage the learned knowledge in image classification tasks in an o.o.d. setting. With a new training methodology based on synthetic datasets that are constructed under the causal framework, the deep neural networks acquire knowledge from semantically different domains (e.g. even from noise), and exhibit certain level of systematicity in parameter estimation experiments. Based on this, a novel architecture is devised consisting of a classifier, an estimator and an identifier (abbreviated as "CED"). By emulating the "hypothesis-verification" process in human visual perception, CED improves the classification accuracy significantly on test sets under covariate shift.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 27 of its 80 references that are in the map. First posted on arXiv 2 Jun 2022.
Builds on
27 of 80 references.
References · 27
- 2014Explaining and Harnessing Adversarial Examplescited by 7,287
- 2014Conditional Generative Adversarial Netscited by 2,152
- 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
- 2015Spatial Transformer Networkscited by 1,423
- 2017Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real Worldcited by 1,401
- 2017Dynamic Routing Between Capsulescited by 689
- 2019DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusioncited by 352
- 2021Graph Neural Network-Based Anomaly Detection in Multivariate Time Seriescited by 300
- 2019Compositionality decomposed: how do neural networks generalise?cited by 264
- 2020Weakly-Supervised Disentanglement Without Compromisescited by 237
- 20203D Object Detection with Pointformercited by 194
- 2016DeepCoder: Learning to Write Programscited by 145
15 more
- 2018Strike (with) a Pose: Neural Networks Are Easily Fooled by Strange Poses of Familiar Objectscited by 143
- 2019Compositional generalization through meta sequence-to-sequence learningcited by 98
- 2020Improving data-driven global weather prediction using deep convolutional neural networks on a cubed spherecited by 95
- 2017Measuring the tendency of CNNs to Learn Surface Statistical Regularitiescited by 87
- 2019AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transformations rather than Datacited by 77
- 2021Towards Learning Universal Audio Representationscited by 51
- 2021VoxelTrack: Multi-Person 3D Human Pose Estimation and Tracking in the Wildcited by 39
- 2018Domain Randomization for Scene-Specific Car Detection and Pose Estimationcited by 23
- 2022Using Natural Language and Program Abstractions to Instill Human Inductive Biases in Machinescited by 20
- 2019Learning Neurosymbolic Generative Models via Program Synthesiscited by 11
- 2019Domain Randomization for Active Pose Estimationcited by 11
- 2019EnAET: A Self-Trained framework for Semi-Supervised and Supervised Learning with Ensemble Transformationscited by 11
- 2019Analysing Deep Reinforcement Learning Agents Trained with Domain Randomisationcited by 9
- 2019Deep Learning with ConvNET Predicts Imagery Tasks Through EEGcited by 2
- 2021Pre-training of Deep RL Agents for Improved Learning under Domain Randomizationcited by 2
Built on by
Citing papers per year: 2023: 1.
1 paper.