Variational Autoencoders with Riemannian Brownian Motion Priors

Authors: Dimitrios Kalatzis, David Eklund, Georgios Arvanitidis, Soren Hauberg

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental For our first experiment we train a VAE with a Riemannian Brownian motion prior (R-VAE) for different latent dimensions and compare it to a VAE with a standard Normal prior and a VAE with a Vamp Prior. Tables 1 & 2 show the results. We next assess the usefulness of the latent representations of R-VAE. R-VAE has produced more separable clusters in the latent space... We quantitatively measured the utility of the R-VAE latent codes in different dimensionalities by training a classifier to predict digit labels and measuring the average overall and per-digit F1 score. Table 3 shows the results when comparing against the same classifier trained on latent codes derived by a VAE.
Researcher Affiliation Academia 1Section for Cognitive Systems, Department of Applied Mathematics and Computer Science, Technical University of Denmark 2Research Institutes of Sweden, Isafjordsgatan 22, 164 40 Kista, Sweden 3Empirical Inference Department, Max Planck Institute for Intelligent Systems, T ubingen, Germany.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it state that code is released or available.
Open Datasets Yes a VAE is trained on the 1-digits of MNIST (Page 2) and Results on Fashion MNIST (Table 2).
Dataset Splits No The paper mentions training models for a certain number of epochs and optimization strategies, but does not specify explicit training, validation, or test dataset splits (e.g., percentages or sample counts).
Hardware Specification No The acknowledgements section states 'We gratefully acknowledge the support of the NVIDIA Corporation with the donation of GPU hardware.' However, it does not specify any particular GPU model or other hardware components.
Software Dependencies No The paper mentions using the 'Adam optimizer (Kingma & Ba, 2015)' but does not provide specific version numbers for this or any other software dependencies.
Experiment Setup Yes All models were trained for 300 epochs. ...All experiments were run with the Adam optimizer (Kingma & Ba, 2015) with default parameter settings and a fixed learning rate of 10-3. The batch size was 100 for all models.