Research map / Image generation / Autoregressive and latent generation
Generative Neurosymbolic Machines
Jindong Jiang, Sungjin Ahn
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 |
|---|---|---|---|
| Use Gaussian global prior instead of StructDRAW (ablation) | GNM · MNIST-10 | S-Acc: 0.824 -> 0.096 | Hurt |
| Use Gaussian global prior instead of StructDRAW (ablation) | GNM · ARROW | S-Acc: 0.976 -> 0.784 | Hurt |
| Remove MLP interaction layer from StructDRAW (ablation) | GNM · MNIST-10 | S-Acc: 0.824 -> 0.128 | Hurt |
| Remove MLP interaction layer from StructDRAW (ablation) | GNM · ARROW | D-Steps: 11099 -> 8799 | Hurt |
| Remove MLP interaction layer from StructDRAW (ablation) | GNM · ARROW | log-likelihood: 33809 -> 33812 | No clear effect |
| Add MLP interaction layer to ConvDRAW (ablation) | ConvDRAW · ARROW | S-Acc: 0.176 -> 0.844 | Improved |
| Add MLP interaction layer to ConvDRAW (ablation) | ConvDRAW · MNIST-10 | S-Acc: 0.000 -> 0.104 | Mixed |
| Add MLP interaction layer to ConvDRAW (ablation) | ConvDRAW · ARROW | D-Steps: 3800 -> 2799 | Hurt |
| Increase ConvDRAW drawing steps from 4 to 8 | ConvDRAW · ARROW | S-Acc: 0.176 -> 0.420 | Improved |
| Increase ConvDRAW drawing steps from 4 to 8 | ConvDRAW · MNIST-10 | S-Acc: 0.000 -> 0.000 | No clear effect |
| Increase GNM drawing steps from 4 to 8 | GNM · MNIST-4-10 | S-Acc: 0.692 -> 0.852 | Improved |
| Increase beta of KL term to 2 for GNM | GNM · MNIST-4-10 | S-Acc: 0.692 -> 0.916 | Improved |
Closest papers by meaning
- Neurosymbolic Grounding for Compositional World Models (ICLR 2024 · similarity 0.86)
- Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding (ICML 2024 · similarity 0.84)
- Symbolic Representation for Any-to-Any Generative Tasks (CVPR 2025 · similarity 0.84)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.84)
- Hyper-Transforming Latent Diffusion Models (ICML 2025 · similarity 0.84)
- DDMI: Domain-agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations (ICLR 2024 · 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.
