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 (15)
- Deep Learning in Neural Networks: An Overview
- Shortcut Learning in Deep Neural Networks
- Deep Learning: A Critical Appraisal
- Weakly-Supervised Disentanglement Without Compromises
- 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
- 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