Research map / Learning theory and algorithms / Sample complexity
Divergence Frontiers for Generative Models: Sample Complexity, Quantization Effects, and Frontier Integrals
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 smoothed distribution estimators (Good-Turing, Krichevsky-Trofimov, Laplace, Braess-Sauer) instead of empirical plug-in | frontier integral estimator · synthetic and real data (CIFAR-10, WikiText-103) | absolute error | Helped |
| Good-Turing estimator across tail-decay regimes | smoothed estimators · synthetic Zipf(r) pairs | absolute error | Helped |
| Krichevsky-Trofimov and Braess-Sauer estimators | smoothed estimators · synthetic data, varying tail decay | absolute error | Mixed |
| smoothing when k/n is large | frontier integral estimator · synthetic and real data | absolute error | Helped |
| set quantization level k = n^(1/3) rather than n^(1/2), n^(1/4), n^(1/5) (ablation) | quantized estimator · 2-d Gaussian and t-distributions | total absolute error: r=3 best for n >= 10^4 | Helped |
| vary r among 3, 4, 5 at small n (ablation) | quantized estimator · 2-d Gaussian and t-distributions | total absolute error | No clear effect |
| distribution-independent bound vs oracle bound for tracking error slope | statistical error bound · synthetic data, varying k | slope match: oracle correct for k > 300 | Mixed |
| oracle bound vs distribution-independent bound on real data | statistical error bound · WikiText-103 and CIFAR-10 | slope match | Mixed |
Closest papers by meaning
- Feature Likelihood Divergence: Evaluating the Generalization of Generative Models Using Samples (NeurIPS 2023 · similarity 0.90)
- Assessing Generative Models via Precision and Recall (NeurIPS 2018 · similarity 0.89)
- DiverGen: Improving Instance Segmentation by Learning Wider Data Distribution with More Diverse Generative Data (CVPR 2024 · similarity 0.86)
- Image Generation Diversity Issues and How to Tame Them (CVPR 2025 · similarity 0.86)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.86)
- A Likelihood Based Approach to Distribution Regression Using Conditional Deep Generative Models (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.