Research map / Probabilistic and scientific modeling / Variational and generative sampling
Emergent Asymmetry of Precision and Recall for Measuring Fidelity and Diversity of Generative Models in High Dimensions
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 |
|---|---|---|---|
| Increase number of dimensions d for hyperspherical generated support at fixed radius (moving generated support outside vs inside real support) | Precision and Recall (k-NN based) · Synthetic hyperspheres | Precision/Recall asymmetry | Hurt |
| Replace Precision and Recall with symmetric versions taking the minimum with complement metrics (symPrecision, symRecall) | Precision and Recall · Synthetic hyperspheres and CelebA/CIFAR10 feature-space scaling and contrast experiments | symmetry of metric vs. support scaling/contrast factor | Helped |
| Use contrast-scaled generators (s=0.4 washed-out vs s>=1) and compare Precision/Recall | Precision and Recall · CelebA and CIFAR10 varying contrast | Precision and Recall | Hurt |
| Increase number of samples at fixed high dimension (d>64) for hypersphere experiments (ablation) | Precision and Recall · Synthetic hyperspheres | Precision/Recall asymmetry | No clear effect |
| Use random R64 embedding instead of VGG16 for CelebA and CIFAR10 experiments (ablation) | Precision and Recall and symmetric variants · CelebA and CIFAR10 with random embedding | asymmetry pattern | No clear effect |
| Change K of k-NN (K=3 vs K=5) for the manifold approximation (ablation) | Precision and Recall · CelebA and CIFAR10 experiments | reported asymmetry behavior | No clear effect |
Closest papers by meaning
- A Unifying Information-theoretic Perspective on Evaluating Generative Models (AAAI 2025 · similarity 0.92)
- Assessing Generative Models via Precision and Recall (NeurIPS 2018 · similarity 0.91)
- Position: All Current Generative Fidelity and Diversity Metrics are Flawed (ICML 2025 · similarity 0.90)
- Efficient Precision and Recall Metrics for Assessing Generative Models using Hubness-aware Sampling (ICML 2024 · similarity 0.89)
- Metric Space Magnitude for Evaluating the Diversity of Latent Representations (NeurIPS 2024 · similarity 0.87)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.87)
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.