Identifying Metric Structures of Deep Latent Variable Models

Authors: Stas Syrota, Yevgen Zainchkovskyy, Johnny Xi, Benjamin Bloem-Reddy, Søren Hauberg

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate that our theory results in more reliable latent distances, offering a principled path forward in extracting trustworthy conclusions from deep latent variable models.
Researcher Affiliation Academia 1Department of Applied Mathematics and Computer Science, Technical University of Denmark, Lyngby, Denmark 2Department of Statistics, University of British Columbia. Correspondence to: Stas Syrota <EMAIL>.
Pseudocode No The paper describes the methodology for computing geodesics in Appendix C but does not present it in a structured pseudocode or algorithm block.
Open Source Code Yes The code to reproduce our results is available in the project repository Git Hub1. 1https://github.com/mustass/identifiable-latent-metric-space
Open Datasets Yes We train this model on a 3-class subset of MNIST (Deng, 2012) with a 2D latent space for visualization purposes and full CIFAR10 (Krizhevsky et al.).
Dataset Splits No The paper mentions using a 'test set' for evaluating distances but does not provide specific percentages, sample counts, or explicit methodology for training, validation, and test dataset splits.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions optimizers like Adam and specific model architectures like M-flows and RQS splines, but it does not provide specific version numbers for any software dependencies or libraries used for implementation.
Experiment Setup Yes We train 30 models with different initial seeds and compute both Euclidean (d E) and geodesic (dg) distances in the latent space between 100 randomly chosen unique point pairs...