Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Kernelised Stein Statistic for Assessing Implicit Generative Models
Wenkai Xu, Gesine Reinert
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 NP-KSD instead of MMDAgg | NP-KSD · MNIST DCGAN | rejection rate: 0.73 -> 0.92 | Improved |
| Use NP-KSD_m (mean statistic) instead of NP-KSD (ablation) | NP-KSD · MNIST DCGAN | rejection rate: 0.92 -> 1.00 | Improved |
| Use NP-KSD instead of MMDAgg | NP-KSD · MNIST VAE | rejection rate: 0.93 -> 1.00 | Improved |
| Use NP-KSD_m instead of MMDAgg | NP-KSD_m · CIFAR10 DCGAN | rejection rate: 0.48 -> 0.74 | Improved |
| Use NP-KSD_m instead of MMDAgg | NP-KSD_m · CIFAR10 NCSN | rejection rate: 0.57 -> 0.81 | Improved |
| Use NP-KSD_m instead of MMDAgg | NP-KSD_m · CIFAR10.1 | rejection rate: 0.83 -> 0.96 | Improved |
| Increase re-sample size B from 5 to 40 (ablation) | NP-KSD · Runtime study m=40 | rejection rate: 0.24 -> 0.55 | Improved |
| Increase re-sample size B from 5 to 40 (ablation) | NP-KSD · Runtime study m=40 | runtime (s): 4.65 -> 10.44 | Hurt |
| Increase B from 20 to 40 (ablation) | NP-KSD · Runtime study m=40 | rejection rate: 0.51 -> 0.55 | No clear effect |
| Use NP-KSD with B=5 instead of MMDAgg | NP-KSD · Runtime study m=40 | rejection rate: 0.24 -> 0.27 | No clear effect |
| Use misspecified Gaussian conditional model NP-KSD_G (ablation) | NP-KSD · Gaussian Variance Difference | rejection rate | Hurt |
| Use NP-KSD tests instead of KSD and MMDAgg | NP-KSD · Mixture of Gaussians | rejection rate | Improved |
Closest papers by meaning
- Debiasing Synthetic Data Generated by Deep Generative Models (NeurIPS 2024 · similarity 0.88)
- Maximizing the Potential of Synthetic Data: Insights from Random Matrix Theory (ICLR 2025 · similarity 0.87)
- Towards Theoretical Understandings of Self-Consuming Generative Models (ICML 2024 · similarity 0.87)
- Position: All Current Generative Fidelity and Diversity Metrics are Flawed (ICML 2025 · similarity 0.87)
- An Interpretable Evaluation of Entropy-based Novelty of Generative Models (ICML 2024 · similarity 0.86)
- Towards a Scalable Reference-Free Evaluation of Generative Models (NeurIPS 2024 · 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.
