Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VAE Explainer: Supplement Learning Variational Autoencoders with Interactive Visualization

Donald Bertucci, Alex Endert

TL;DR

VAE Explainer is presented, an interactive Variational Autoencoder running in the browser to supplement existing static documentation to supplement existing static documentation on Variational Autoencoders.

Abstract

Variational Autoencoders are widespread in Machine Learning, but are typically explained with dense math notation or static code examples. This paper presents VAE Explainer, an interactive Variational Autoencoder running in the browser to supplement existing static documentation (e.g., Keras Code Examples). VAE Explainer adds interactions to the VAE summary with interactive model inputs, latent space, and output. VAE Explainer connects the high-level understanding with the implementation: annotated code and a live computational graph. The VAE Explainer interactive visualization is live at https://xnought.github.io/vae-explainer and the code is open source at https://github.com/xnought/vae-explainer.

VAE Explainer: Supplement Learning Variational Autoencoders with Interactive Visualization

TL;DR

VAE Explainer is presented, an interactive Variational Autoencoder running in the browser to supplement existing static documentation to supplement existing static documentation on Variational Autoencoders.

Abstract

Variational Autoencoders are widespread in Machine Learning, but are typically explained with dense math notation or static code examples. This paper presents VAE Explainer, an interactive Variational Autoencoder running in the browser to supplement existing static documentation (e.g., Keras Code Examples). VAE Explainer adds interactions to the VAE summary with interactive model inputs, latent space, and output. VAE Explainer connects the high-level understanding with the implementation: annotated code and a live computational graph. The VAE Explainer interactive visualization is live at https://xnought.github.io/vae-explainer and the code is open source at https://github.com/xnought/vae-explainer.
Paper Structure (6 sections, 2 equations, 4 figures)

This paper contains 6 sections, 2 equations, 4 figures.

Table of Contents

  1. Introduction
  2. System
  3. High-Level Summary View
  4. Low-Level Graph View
  5. Implementation
  6. Conclusion

Figures (4)

  • Figure 1: Users can draw a digit as input and the VAE runs in real-time. VAE Explainer displays the encoded distribution on top of the latent space. Then, we sample a point from the distribution and decode into the reconstruction.
  • Figure 2: Users can click "Explain VAE Details" to show annotated code connected to a computational graph. Hovering over lines in the code will highlight portions of the graph.
  • Figure 3: Left side: as we draw the digit "0" in the input, the encoded distribution changes location and size to represent the distribution of possible "0"s. Right side: as we hover the latent space and change the sampled point, we interpolate the reconstruction.
  • Figure 4: VAE sampling Keras code chollet2015keras accompanied by its computational graph. Extra labels have been removed for figure presentation.