Research map / Vision-language and multimodal / Human and embodied modeling
SCAN: Learning Hierarchical Compositional Visual Concepts
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 SCAN's forward-KL grounding with disentangled visual primitives | SCAN · DeepMind Lab test operators | accuracy: 0.67 -> 0.79 | Helped |
| Use SCAN's forward-KL grounding with disentangled visual primitives | SCAN · DeepMind Lab test symbols | diversity: 9.2 -> 1.08 | Helped |
| Replace forward grounding KL with reverse grounding KL (ablation) | SCAN · DeepMind Lab train symbols | accuracy: 0.82 -> 0.86 | Helped |
| Replace forward grounding KL with reverse grounding KL (ablation) | SCAN · DeepMind Lab train symbols | diversity: 1.46 -> 13.17 | Hurt |
| Use unstructured visual representations rather than disentangled visual primitives (ablation) | SCAN · DeepMind Lab test symbols | accuracy: 0.79 -> 0.42 | Hurt |
| Replace learned recombination operators with closed-form operators (ablation) | SCAN recombination module · Recombination operator evaluation | accuracy: 0.79 vs 0.54 | Hurt |
| Replace learned recombination operators with closed-form operators (ablation) | SCAN recombination module · Recombination operator evaluation | diversity: 1.05 vs 2.03 | Hurt |
| Convert DeepMind Lab images from RGB to HSV before β-VAE training (ablation) | SCAN visual representation · DeepMind Lab | disentanglement | Helped |
| Use disentangled rather than entangled visual representations (ablation) | SCAN · dSprites | concept sample quality | Helped |
Closest papers by meaning
- Concept-RuleNet: Grounded Multi-Agent Neurosymbolic Reasoning in Vision Language Models (AAAI 2026 · similarity 0.88)
- Learning Vision and Language Concepts for Controllable Image Generation (ICML 2025 · similarity 0.88)
- Learning to Infer Generative Template Programs for Visual Concepts (ICML 2024 · similarity 0.88)
- Bridging the gap to real-world language-grounded visual concept learning (NeurIPS 2025 · similarity 0.86)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.86)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.86)
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.