Posterior Sampling on Potts q=3, beta=1.2 16x16
12.37Sinkhorn DistanceTB + Buffer + MCMC
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TB + Buffer + MCMCTraining=off-policy, Objective=Trajectory Balance, Buffer=true, MCMC=true2026.02 | 12.37 | 620.73 | 621.3 | 3 | 0 | |
| LV + Buffer + MCMCTraining=off-policy, Objective=Log-Variance, Buffer=true, MCMC=true2026.02 | 19.56 | 619.38 | 621.6 | 13 | 0.02 | |
| LV + BufferTraining=off-policy, Objective=Log-Variance, Buffer=true2026.02 | 32.88 | 619.47 | 628.47 | 21 | 0.02 | |
| MH MCMCType=Sampling-only2026.02 | 47.27 | — | — | 3 | 0.28 | |
| MDNS (WDCE)Loss=WDCE2026.02 | 99.95 | 620.23 | 680.52 | 58 | 0.01 | |
| TB + BufferTraining=off-policy, Objective=Trajectory Balance, Buffer=true2026.02 | 156.12 | 619.84 | 705.56 | 95 | 0 | |
| LVTraining=on-policy, Objective=Log-Variance2026.02 | 156.71 | 619.82 | 748.31 | 95 | 0 | |
| TBTraining=on-policy, Objective=Trajectory Balance2026.02 | 161.78 | 619.34 | 1,117.43 | 96 | 0.01 |