Trajectory Modeling on high-dimensional Gaussian mixture benchmark
0Trajectory reverse KL divergence (D=2, Gaussian noise 0.02)DLightSB
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DLightSB2025.09 | 0 | 0 | 0 | 0 | 5.7 | 0 | 0 | 0 | 0.3 | 0.1 | 0.1 | 0.1 | |
| Feature-wise SB2025.09 | 0.2 | 0.2 | 0 | 0 | 0.2 | 0.5 | 0.3 | 0.3 | 0 | 0.4 | 0.3 | 0.4 | |
| CSBMLoss=KL, N + 1=642025.09 | 0.4 | 1.8 | 0.4 | 0.4 | 2.7 | 7.1 | 10.3 | 10.8 | 22.3 | 2.5 | 11.3 | 10.8 | |
| α-CSBMLoss=MSE, N + 1=642025.09 | 0.5 | 1.1 | 1.4 | 1.3 | 11.4 | 2.8 | 16.9 | 20 | 75.9 | 3.3 | 46.7 | 48.3 | |
| DLightSB-MLoss=KL, N + 1=162025.09 | 0.5 | 0 | 0.1 | 0.2 | 21.7 | 1 | 7.4 | 5.9 | 6.4 | 2.3 | 322.2 | 329.7 | |
| α-CSBMLoss=KL, N + 1=642025.09 | 0.7 | 1.4 | 0.4 | 0.3 | 4.4 | 1 | 7.3 | 10.7 | 22.7 | 3 | 17.1 | 16.2 | |
| DLightSB-MLoss=KL, N + 1=642025.09 | 0.7 | 0.1 | 0.3 | 0.4 | 8.2 | 1.4 | 12.3 | 11.7 | 6.8 | 1.9 | 292.6 | 110.7 | |
| DLightSB-MLoss=MSE, N + 1=162025.09 | 1.1 | 0.3 | 0.3 | 0.2 | 182.3 | 4.5 | 6.5 | 6.6 | 25.4 | 15.8 | 70.3 | 405 | |
| DLightSB-MLoss=MSE, N + 1=642025.09 | 1.4 | 0.5 | 0.4 | 0.2 | 156.4 | 8.2 | 8.4 | 8.9 | 52.4 | 27.6 | 313.8 | 104.1 | |
| CSBMLoss=KL, N + 1=162025.09 | 1.5 | 3.7 | 1 | 1.1 | 7.7 | 22.1 | 50.9 | 71.4 | 15.8 | 13 | 16.6 | 20.2 | |
| α-CSBMLoss=KL, N + 1=162025.09 | 1.7 | 3.3 | 0.8 | 0.9 | 8.1 | 17.5 | 41.8 | 53.8 | 23.6 | 5 | 20.5 | 27.3 | |
| Reference2025.09 | 2.1 | 3 | 0.8 | 1.3 | 2.2 | 4.6 | 1.4 | 1.5 | 1.4 | 3.5 | 1 | 1.2 | |
| α-CSBMLoss=MSE, N + 1=162025.09 | 2.7 | 3 | 1.7 | 1.1 | 9.6 | 18.5 | 52.5 | 75.4 | 54.1 | 5.6 | 83.6 | 89.5 | |
| CSBMLoss=MSE, N + 1=162025.09 | 4.2 | 2.5 | 3.3 | 4.3 | 8.1 | 19.5 | 88.3 | 122.3 | 55.4 | 12.2 | 41.8 | 34.7 | |
| CSBMLoss=MSE, N + 1=642025.09 | 10.2 | 1.8 | 2.1 | 2.2 | 4.4 | 9.1 | 22.2 | 24.4 | 75.6 | 2.6 | 38.5 | 39 |