Research map / Image generation / Autoregressive and latent generation
One-Shot Generalization in Deep Generative Models
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 spatial-transformer attention instead of fully connected attention (ablation) | Sequential generative model · MNIST | test negative log-likelihood: ≤98.7(0.8) -> ≤80.5(0.3) | Helped |
| Use spatial-transformer attention instead of fully connected attention (ablation) | Sequential generative model · Multi-MNIST | test negative log-likelihood: 270.3(0.8) -> 182.0(0.6) | Helped |
| Use additive canvas instead of CGRU canvas with spatial-transformer attention (ablation) | Sequential generative model · Multi-MNIST | test negative log-likelihood: 194.9(0.5) -> 182.0(0.6) | Helped |
| Increase sequence length from 30 to 80 steps (ablation) | Sequential generative model with spatial-transformer attention and CGRU · MNIST | test negative log-likelihood: ≤81.5(0.4) -> ≤80.5(0.3) | Helped |
| Increase sequence length from 30 to 80 steps (ablation) | Sequential generative model with spatial-transformer attention and additive canvas · MNIST | test negative log-likelihood: ≤82.6(0.5) -> ≤81.6(0.4) | Helped |
| Use 80 steps instead of 40 with spatial-transformer attention and additive canvas (ablation) | Sequential generative model · Omniglot 52×52 | test negative log-likelihood: 136.1(0.4) -> 141.5(0.5) | Hurt |
| Share the prior LSTM rather than use a separate recurrent function in inference (ablation) | Sequential generative model · reported performance | performance | No clear effect |
| Use the additive canvas instead of the CGRU canvas with spatial-transformer attention (ablation) | Sequential generative model · Omniglot 52×52 | test negative log-likelihood: 144.5(0.4) -> 141.5(0.5); 134.1(0.5) -> 141.5(0.5) | Mixed |
Closest papers by meaning
- Diversity vs. Recognizability: Human-like generalization in one-shot generative models (NeurIPS 2022 · similarity 0.91)
- Intriguing Properties of Generative Classifiers (ICLR 2024 · similarity 0.87)
- Understanding prompt engineering may not require rethinking generalization (ICLR 2024 · similarity 0.86)
- A Generalization Theory for Zero-Shot Prediction (ICML 2025 · similarity 0.86)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.86)
- No Free Lunch: Fundamental Limits of Learning Non-Hallucinating Generative Models (ICLR 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.