Log Z estimation on MNIST downsampled (test)
0.11Log Z Absolute ErrorMCD
Evaluation Results
| Method | Links | |
|---|---|---|
| MCDDimension=7 × 7, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 0.11 | |
| ULADimension=7 × 7, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 0.14 | |
| UHADimension=7 × 7, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 0.17 | |
| MCDDimension=14 × 14, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 6.25 | |
| ULADimension=14 × 14, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 13.24 | |
| UHADimension=14 × 14, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 15.04 | |
| MCDDimension=28 × 28, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 23.1 | |
| UHADimension=28 × 28, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 82.16 | |
| ULADimension=28 × 28, Importance samples=4096, Training steps=100K, Batch size=128, Base Model=NICE2022.08 | 141.29 |