bikeshc / scipy2015

Talk for SciPy2015 "Deep Learning: Tips From The Road"

PDF version of slides included. Slides also uploaded to https://speakerdeck.com/kastnerkyle

Video of talk here: https://www.youtube.com/watch?v=TBBtOeY2Q78

To run the code in the graphics directory, you will need Theano. I have not run this on CPU yet, but it runs pretty quickly on GPU. To run the code, simply go to graphics_code/vae or graphics_code/cvae

THEANO_FLAGS="floatX=float32,device=gpu,mode=FAST_RUN" python vae.py

or

THEANO_FLAGS="floatX=float32,device=gpu,mode=FAST_RUN" python cvae.py

will start training the model. After training,

THEANO_FLAGS="floatX=float32,device=gpu,mode=FAST_RUN" python flying_vae.py serialized_vae.pkl

or

THEANO_FLAGS="floatX=float32,device=gpu,mode=FAST_RUN" python flying_cvae.py serialized_cvae.pkl

will generate plots for the saved model.

Variational Autoencoder

This variational autoencoder follows the general procedure described in Auto-Encoding Variational Bayes, Kingma and Welling

Another paper describes a similar concept, Stochastic Backpropagation and Approximate Inference in Deep Generative Models, Rezende, Mohamed, and Wierstra.

VAE Code Walking

VAE Reconstruction

Conditional Variational Autoencoder

This conditional variational autoencoder follows a similar procedure to that described in Semi-supervised Learning with Deep Generative Models, Kingma, Rezende, Mohamed, and Welling.

Conditional VAE Code Walking With Conditional Control

Holding Style (Z) Fixed and Changing Conditional y

Conditional VAE Reconstruction and Prediction

Linked content

sklearn-theano, a scikit-learn compatible library for using pretrained networks http://sklearn-theano.github.io/

My research code https://github.com/kastnerkyle/santa_barbaria

Neural network tutorial by @NewMu / Alec Radford https://github.com/Newmu/Theano-Tutorials

Theano Deep Learning Tutorials http://deeplearning.net/tutorial/