Research map / Vision-language and multimodal / Human and embodied modeling
Are Disentangled Representations Helpful for Abstract Visual Reasoning?
Sjoerd van Steenkiste, Francesco Locatello, Jürgen Schmidhuber, Olivier Bachem
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use more disentangled pretrained representations instead of less disentangled representations | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning | downstream accuracy in the few-sample regime | Improved |
| Use representations with higher FactorVAE scores | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, up to 20K steps | downstream accuracy | Improved |
| Use representations with higher BetaVAE scores | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, up to 20K steps | downstream accuracy | Improved |
| Train with more labeled samples, from the few-sample to the many-sample regime | WReN using disentangled representations · dSprites and 3dshapes abstract visual reasoning | correlation between disentanglement and downstream accuracy | Hurt |
| Use higher reconstruction quality instead of higher disentanglement | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, few-sample regime | correlation with downstream accuracy | Hurt |
| Use higher reconstruction quality | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, many-sample regime | correlation with downstream accuracy | Improved |
| Use representations ranked by DCI Disentanglement rather than by FactorVAE or BetaVAE score | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, few-sample regime | correlation with downstream accuracy | Hurt |
| Use representations ranked by MIG or SAP rather than by FactorVAE or BetaVAE score | WReN abstract reasoning models · dSprites and 3dshapes abstract visual reasoning, few-sample regime | correlation with downstream accuracy | Hurt |
Closest papers by meaning
- Revisiting Disentanglement in Downstream Tasks: A Study on Its Necessity for Abstract Visual Reasoning (AAAI 2024 · similarity 0.93)
- Beyond the Doors of Perception: Vision Transformers Represent Relations Between Objects (NeurIPS 2024 · similarity 0.86)
- Slot Abstractors: Toward Scalable Abstract Visual Reasoning (ICML 2024 · similarity 0.85)
- MARVEL: Multidimensional Abstraction and Reasoning through Visual Evaluation and Learning (NeurIPS 2024 · similarity 0.84)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.84)
- ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
