Visual Representation Learning Does Not Generalize Strongly Within the Same Domain
arXiv:2107.08221
Abstract
An important component for generalization in machine learning is to uncover underlying latent factors of variation as well as the mechanism through which each factor acts in the world. In this paper, we test whether 17 unsupervised, weakly supervised, and fully supervised representation learning approaches correctly infer the generative factors of variation in simple datasets (dSprites, Shapes3D, MPI3D) from controlled environments, and on our contributed CelebGlow dataset. In contrast to prior robustness work that introduces novel factors of variation during test time, such as blur or other (un)structured noise, we here recompose, interpolate, or extrapolate only existing factors of variation from the training data set (e.g., small and medium-sized objects during training and large objects during testing). Models that learn the correct mechanism should be able to generalize to this benchmark. In total, we train and test 2000+ models and observe that all of them struggle to learn the underlying mechanism regardless of supervision signal and architectural bias. Moreover, the generalization capabilities of all tested models drop significantly as we move from artificial datasets towards more realistic real-world datasets. Despite their inability to identify the correct mechanism, the models are quite modular as their ability to infer other in-distribution factors remains fairly stable, providing only a single factor is out-of-distribution. These results point to an important yet understudied problem of learning mechanistic models of observations that can facilitate generalization.
References in corpus (34)
- Deep Learning in Neural Networks: An Overview
- Auto-Encoding Variational Bayes
- Shortcut Learning in Deep Neural Networks
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
- Disentangling by Factorising
- Group Equivariant Convolutional Networks
- Learning Representations and Generative Models for 3D Point Clouds
- Deep Learning: A Critical Appraisal
- Invariant Risk Minimization
- Understanding disentangling in -VAE
- Unsupervised Feature Extraction by Time-Contrastive Learning and Nonlinear ICA
- Invariant Models for Causal Transfer Learning
- Measuring abstract reasoning in neural networks
- Benchmarking Robustness in Object Detection: Autonomous Driving when Winter is Coming
- General -Equivariant Steerable CNNs
- Weakly-Supervised Disentanglement Without Compromises
- Learning Independent Causal Mechanisms
- Weakly Supervised Disentanglement with Guarantees
- Deep Set Prediction Networks
- On the Transfer of Inductive Bias from Simulation to the Real World: a New Disentanglement Dataset
- Effects of Degradations on Deep Neural Network Architectures
- On the Binding Problem in Artificial Neural Networks
- On Disentangled Representations Learned From Correlated Data
- ICE-BeeM: Identifiable Conditional Energy-Based Deep Models Based on Nonlinear ICA
- Nonlinear Invariant Risk Minimization: A Causal Approach
- Generalization in anti-causal learning
- Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse Coding
- On Linear Identifiability of Learned Representations
- On the Transfer of Disentangled Representations in Realistic Settings
- How intelligent are convolutional neural networks?
- Do Neural Networks for Segmentation Understand Insideness?
- A theory of independent mechanisms for extrapolation in generative models
- Clusterability in Neural Networks
- Challenge of Spatial Cognition for Deep Learning