Image Discovery on MNIST (val)
0.988SSIMJADAI
Evaluation results
| Method | Links | ||
|---|---|---|---|
| JADAIsigma=0, policy=learned, backbone=Flow Matching (FM), terminal_horizon=T=62025.12 | 0.988 | 0.185 | |
| JADAIsigma=0.001, policy=learned, backbone=Flow Matching (FM), terminal_horizon=T=62025.12 | 0.985 | 0.218 | |
| JADAIsigma=0.01, policy=learned, backbone=Flow Matching (FM), terminal_horizon=T=62025.12 | 0.981 | 0.249 | |
| JADAIsigma=0, policy=learned, backbone=Diffusion Models (DM), terminal_horizon=T=62025.12 | 0.968 | 0.177 | |
| JADAIsigma=0.001, policy=learned, backbone=Diffusion Models (DM), terminal_horizon=T=62025.12 | 0.966 | 0.208 | |
| JADAIsigma=0.01, policy=learned, backbone=Diffusion Models (DM), terminal_horizon=T=62025.12 | 0.96 | 0.246 | |
| CoDiffsigma=0, policy=learned, terminal_horizon=T=62025.12 | 0.826 | — | |
| JADAIsigma=0, policy=random, backbone=Diffusion Models (DM), terminal_horizon=T=62025.12 | 0.478 | 2.056 | |
| CoDiffsigma=0, policy=random, terminal_horizon=T=62025.12 | 0.463 | — | |
| JADAIsigma=0, policy=random, backbone=Flow Matching (FM), terminal_horizon=T=62025.12 | 0.451 | 2.168 |