How to Generate Images – Intro to Deep Learning #14


We’re going to build a variational autoencoder capable of generating novel images after being trained on a collection of images. We’ll be using handwritten digit images as training data. Then we’ll both generate new digits and plot out the learned embeddings. And I introduce Bayesian theory for the first time in this series 🙂

Code for this video:
https://github.com/llSourcell/how_to_generate_images

Mike’s Winning Code:
https://github.com/xkortex/how_to_win_slot_machines/blob/master/WallStBandits.ipynb

SG’s Runner up Code:
https://github.com/esha-sg/Intro-DeepLearning-Siraj-Week13

Please subscribe! And like. And comment. That’s what keeps me going.

2 things
-The embedding visualization at the end would be more spread out if i trained it for more epochs (50 is recommended) but i just used 5.
-The code in the video doesn’t fully implement the reparameterization trick (to save space) but check the GitHub repo for details on that.

More Learning resources:
https://jaan.io/what-is-variational-autoencoder-vae-tutorial/
http://kvfrans.com/variational-autoencoders-explained/
http://blog.fastforwardlabs.com/2016/08/12/introducing-variational-autoencoders-in-prose-and.html
http://blog.fastforwardlabs.com/2016/08/22/under-the-hood-of-the-variational-autoencoder-in.html
http://blog.evjang.com/2016/11/tutorial-categorical-variational.html
https://jmetzen.github.io/2015-11-27/vae.html

Join us in the Wizards Slack channel:
http://wizards.herokuapp.com/

And please support me on Patreon:
https://www.patreon.com/user?u=3191693

Video “How to Generate Images – Intro to Deep Learning #14” Author: Siraj Raval