Multi-Level Variational Autoencoder: Learning Disentangled Representations From Grouped Observations

Authors: Diane Bouchacourt, Ryota Tomioka, Sebastian Nowozin

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally show that our model (i) learns a semantically meaningful disentanglement, (ii) enables control over the latent representation, and (iii) generalises to unseen groups.
Researcher Affiliation Collaboration Diane Bouchacourt OVAL Group University of Oxford diane@robots.ox.ac.uk Ryota Tomioka, Sebastian Nowozin Machine Intelligence and Perception Group Microsoft Research Cambridge, UK {ryoto,Sebastian.Nowozin}@microsoft.com
Pseudocode Yes Algorithm 1: ML-VAE training algorithm.
Open Source Code No The paper does not contain an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We perform evaluation on MNIST (Le Cun et al. 1998). ... Next, we perform evaluation on the face aligned version of the MS-Celeb-1M data set (Guo et al. 2016).
Dataset Splits Yes We randomly separate the 60, 000 training examples into 50, 000 training samples and 10, 000 validation samples, and use the standard MNIST testing data set.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models, processor types, or memory amounts used for running experiments.
Software Dependencies No The paper does not specify particular software dependencies with version numbers (e.g., programming languages or libraries like Python 3.x, PyTorch 1.x).
Experiment Setup Yes The style and content vectors are of size 10 each. The decoder network is composed a linear layer with 500 hidden units with the hyperbolic tangent activation function. It is followed by two linear layers of 784 hidden units each that output respectively the mean and log-variance of p(xi|c G,i, si; θ).