Intrinsic dimensionality estimation using Normalizing Flows

Authors: Christian Horvat, Jean-Pascal Pfister

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our method on various datasets, including 64x64 RGB images, where we achieve state-of-the-art results. We benchmark our method with two NN and LIDL, see Section 3. The latter is very similar to our method as we both inflate the manifold with different values of σ2 and then use an NF to learn qσ2.
Researcher Affiliation Academia Christian Horvat Department of Physiology University of Bern christian.horvat@unibe.ch Jean-Pascal Pfister Department of Physiology Bern, Switzerland jeanpascal.pfister@unibe.ch
Pseudocode Yes Algorithm 4: Estimating the intrinsic dimensionality given a set of data points.
Open Source Code Yes The code for using ID-NF or reproducing our experiments can be found here https://github.com/chrvt/ID-NF.
Open Datasets Yes We test our method on various synthetic datasets with known ID: a sphere, torus, hyperboloid, thin spiral, swiss roll, and Stiefel manifold, see Table 1. We apply our method on the Style Gan d = 2 and d = 64 image manifold...FFHQ dataset [18]. Celeb A dataset [24].
Dataset Splits No The paper mentions using training sets and implicitly testing, but does not specify clear train/validation/test splits with percentages, counts, or explicit standard split references.
Hardware Specification No Did you include the total amount of comute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No]
Software Dependencies No The paper mentions training details are in the supplemental material, but the main text does not provide specific software names with version numbers for reproducibility.
Experiment Setup No We refer to the supplementary for corresponding training details and additional figures.