Intrinsic Dimension Estimation Using Wasserstein Distances
About
It has long been thought that high-dimensional data encountered in many practical machine learning tasks have low-dimensional structure, i.e., the manifold hypothesis holds. A natural question, thus, is to estimate the intrinsic dimension of a given population distribution from a finite sample. We introduce a new estimator of the intrinsic dimension and provide finite sample, non-asymptotic guarantees. We then apply our techniques to get new sample complexity bounds for Generative Adversarial Networks (GANs) depending only on the intrinsic dimension of the data.
Adam Block, Zeyu Jia, Yury Polyanskiy, Alexander Rakhlin• 2021
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Intrinsic Dimension Estimation | MNIST | Intrinsic Dimension Estimate9.49 | 13 | |
| Intrinsic Dimension Estimation | M5 manifold d=2 | Mean Dimension Estimate2.35 | 10 | |
| Intrinsic Dimension Estimation | M8 manifold d=2 | Mean Dimension Estimate2.29 | 10 | |
| Intrinsic Dimension Estimation | M10 Manifold (d=2) | Estimated Dimension2.44 | 10 | |
| Intrinsic Dimension Estimation | Manifold M5 (d=2) n=2000 (uniform samples) | Mean Dimension Estimate2.26 | 10 | |
| Intrinsic Dimension Estimation | Manifold M7 (d=2) n=2000 (uniform samples) | Mean Dimension Estimate2.44 | 10 | |
| Intrinsic Dimension Estimation | Manifold M9 d=2 n=2000 (uniform samples) | Mean Dimension Estimate2.29 | 10 | |
| Intrinsic Dimension Estimation | Manifold M10 d=2 n=2000 (uniform samples) | Mean Dimension Estimate2.31 | 10 | |
| Intrinsic Dimension Estimation | MNL5(12) (n=2000) | Mean Dimension Estimate5.8 | 10 | |
| Intrinsic Dimension Estimation | M6 manifold d=1 | Mean Dimension Estimate2.21 | 10 |
Showing 10 of 54 rows