Cross-Reynolds prediction on 2D Navier-Stokes out-of-distribution Re=10,000 (test)
61.08OOD AccuracyPersistence
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PersistenceParams=0, Type=baseline, Representation=–, Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 61.08 | — | |
| FNO (3-seed)Params=466K, Type=learned predictor, Representation=global spectral, Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 46.68 | 0.11 | |
| PCA relayParams=0, Type=retrieval, no learned dynamics, Representation=global linear (36-d), Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 41.77 | — | |
| kNN-copyParams=0, Type=retrieval, no learned dynamics, Representation=raw field (4,096-d), Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 41.01 | — | |
| ConvAE-Relay (3-seed)Params=1.6M, Type=retrieval, source-trained, Representation=learned local (64-d), Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 38.34 | 0.07 | |
| U-Net (3-seed)Params=7.8M, Type=learned predictor, Representation=local multi-scale, Training regime=Re≈1,000, Rollout steps=10-step autoregressive, N=5002026.05 | 34.72 | 0.6 |