Table 1. Synthetic benchmark datasets.
The synthetic benchmark datasets used for comparison are the first 15 manifolds from Campadelli et al. (2015). The datasets represent various types of manifolds with or without curvature, also with uniform or non-uniform sampling of n = 2, 500 points.
| Dataset | Description | d | embed-dim | |
|---|---|---|---|---|
| 1 | M 1 | 10d sphere | 10 | 11 |
| 2 | M 2 | 3d affine space | 3 | 5 |
| 3 | M 3 | 4 figure | 4 | 6 |
| 4 | M 4 | 4d manifold in 8d | 4 | 8 |
| 5 | M 5 | 2d helix in 3d | 2 | 3 |
| 6 | M 6 | 6dim manifold in 36d | 6 | 36 |
| 7 | M 7 | swiss roll | 2 | 3 |
| 8 | M 9 | 20d affine space | 20 | 20 |
| 9 | M 10a | 10d hypercube | 10 | 10 |
| 10 | M 10b | 17d hypercube | 17 | 17 |
| 11 | M 10c | 24d hypercube | 24 | 24 |
| 12 | M 10d | 70d hypercube | 70 | 70 |
| 13 | M 11 | Moebius band 10x twisted | 2 | 3 |
| 14 | M 12 | Multivariate Gaussian | 20 | 20 |
| 15 | M 13 | 1d curve in 13d | 1 | 13 |