Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Drawing out of Distribution with Neuro-Symbolic Generative Models
Yichao Liang, Joshua B. Tenenbaum, Tuan Anh Le, Siddharth N
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| replace differentiable spline renderer with a neural decoder (remove explicit spline decoder) (ablation) | DooD · out-of-distribution reconstruction (trained on MNIST) | reconstruction quality | Hurt |
| remove execution-guided inference (no canvas-so-far conditioning) (ablation) | DooD · out-of-distribution reconstruction (trained on MNIST) | stroke-count selectivity and reconstruction | Hurt |
| use DooD instead of AIR | stroke-based generative model · Omniglot 20-way one-shot classification | accuracy: 14.5\% -> 73.5\% | Improved |
| use DooD instead of GNS | unconditional generator · Omniglot unconditional generation | FID: 0.133 -> 0.051 | Improved |
| use DooD stroke latents instead of AIR part representations | k-means clustering of inferred strokes · stroke clustering | Silhouette Coefficient: 0.11 -> 0.21 | Improved |
| train on MNIST or Omniglot as source dataset instead of other datasets | DooD and AIR · zero-shot transfer log marginal likelihood | log marginal likelihood | Hurt |
| train and test on same dataset (diagonal) instead of transferring | DooD and AIR · zero-shot transfer log marginal likelihood | log marginal likelihood | Improved |
| use Gaussian instead of Laplace likelihood for Omniglot-trained AIR | AIR · Omniglot-trained transfer | reconstruction | Improved |
Closest papers by meaning
- Latent Representation Matters: Human-like Sketches in One-shot Drawing Tasks (NeurIPS 2024 · similarity 0.87)
- Geometry of naturalistic object representations in recurrent neural network models of working memory (NeurIPS 2024 · similarity 0.85)
- What Sketch Explainability Really Means for Downstream Tasks? (CVPR 2024 · similarity 0.85)
- Evaluating alignment between humans and neural network representations in image-based learning tasks (NeurIPS 2024 · similarity 0.85)
- SketchINR: A First Look into Sketches as Implicit Neural Representations (CVPR 2024 · similarity 0.85)
- CTSketch: Compositional Tensor Sketching for Scalable Neurosymbolic Learning (NeurIPS 2025 · similarity 0.85)
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.
