Class-conditional posterior sampling on MNIST
0.014FIDFVD
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FVDNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.014 | 0.077 | 0.012 | 0.467 | |
| DTSNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.019 | 0.1 | 0.02 | 0.494 | |
| DASNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.026 | 0.097 | 0.017 | 0.457 | |
| FKNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.034 | 0.148 | 0.025 | 0.458 | |
| TDSNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.092 | 0.428 | 0.012 | 0.429 | |
| DPSNFEs=10^6, Number of Seeds=5, Generated Samples=5000, Final Sampling Strategy=Best performance under either reward-weighted or uniform final sampling2026.04 | 0.915 | 0.343 | 0.16 | 0.486 |