pith. sign in

arxiv: 2107.08221 · v4 · pith:IZAX6M4Znew · submitted 2021-07-17 · 💻 cs.LG · cs.CV

Visual Representation Learning Does Not Generalize Strongly Within the Same Domain

classification 💻 cs.LG cs.CV
keywords factorsmodelslearningmechanismvariationdatasetsduringgeneralization
0
0 comments X
read the original 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.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. How can embedding models bind concepts?

    cs.CV 2026-05 unverdicted novelty 7.0

    CLIP relies on high-complexity additive binding that prevents generalization to unseen concept combinations, whereas transformers trained from scratch develop low-complexity multiplicative binding functions that enabl...

  2. Learning to Theorize the World from Observation

    cs.LG 2026-05 unverdicted novelty 7.0

    NEO is a probabilistic neural model that induces compositional programs as a learned Language of Thought from non-textual observations and executes them via a shared transition model to enable explanation-driven gener...

  3. Compositionality Emerges in a Narrow Depth-Connectivity Regime: Architecture Constraints and Solution Manifolds

    cs.LG 2026-06 unverdicted novelty 6.0

    Compositionality emerges in neural networks only in a narrow depth-connectivity regime, with gradient descent converging to fractured solutions outside it.

  4. A Systematic Study of Behavioral Cloning for Scientific Data Annotation

    cs.HC 2026-05 unverdicted novelty 6.0

    Introduces 9 synthetic annotation tasks and benchmarks for behavioral cloning, finding hierarchical skill learning, scaling benefits, effective multi-task pretraining, and shared internal representations of task phase...

  5. Learning to Theorize the World from Observation

    cs.LG 2026-05 unverdicted novelty 6.0

    NEO induces compositional latent programs as world theories from observations and executes them to enable explanation-driven generalization.