Autoregressive rollout on CRP2D four-quadrant Riemann (test)
1.23Final Time Relative L2 ErrorMemory-Conditioned Flow-Matching
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Memory-Conditioned Flow-MatchingTrain=Regular, Sampling=U0.05 (10), Context=Uniform delta_t=0.05, 10 steps2026.02 | 1.23 | 2.3 | |
| Memoryless BaselineTrain=Regular, Sampling=U0.05 (10), Context=Uniform delta_t=0.05, 10 steps2026.02 | 1.26 | — | |
| Memory-Conditioned Flow-MatchingTrain=Regular, Sampling=Inc (10), Context=Increasing delta_t, 10 steps2026.02 | 1.29 | 1 | |
| Memoryless BaselineTrain=Regular, Sampling=Inc (10), Context=Increasing delta_t, 10 steps2026.02 | 1.3 | — | |
| Memory-Conditioned Flow-MatchingTrain=Regular, Sampling=S0.10 (10), Context=Sparse delta_t=0.10, 10 steps2026.02 | 1.31 | 2.5 | |
| Memoryless BaselineTrain=Regular, Sampling=S0.10 (10), Context=Sparse delta_t=0.10, 10 steps2026.02 | 1.34 | — | |
| Memory-Conditioned Flow-MatchingTrain=Regular, Sampling=D0.05 (20), Context=Dense delta_t=0.05, 20 steps2026.02 | 1.52 | 4.7 | |
| Memoryless BaselineTrain=Regular, Sampling=D0.05 (20), Context=Dense delta_t=0.05, 20 steps2026.02 | 1.6 | — |