Fun with Generative Models
Organised by Iliana Peneva
Abstract
Neural networks have been used for almost everything: to train a robot how to play football; to generate April Fools jokes/names of metal bands/traditional Irish music/candy heart messages, to distinguish between chihuahua and muffin, to generate cats from drawings. Inspired by Rob's entertaining WARP from 2 years ago and by Janelle Shane's brilliant blog, my proposed project aims to add to this list of fun NNs applications by generating new dog/cat breeds, DnD spells, Pokemons, cake recipes, talk abstracts, news items.
We can decide what to generate at the retreat (there can be people working on different topics)
Suggestions are welcome for other topics :-)
Aims and Objectives
- learn more about different types of neural networks, their capabilities and limitations
- get experience using neural networks in Python/R
Of Interest to
Anyone interested in deep learning (or likes cats/dogs/cakes/DnD/Pokemon)
Resources Necessary
The datasets will be collected and made available before the retreat.
if using Python - Keras and Tensorflow; if using R - neuralnet, or Keras and Tensorflow for R
References
(most relevant)
Andrey Karpathy's excellent blogpost with examples: http://karpathy.github.io/2015/05/21/rnn-effectiveness/
Christopher Olah's blogpost: http://colah.github.io/posts/2015-08-Understanding-LSTMs/
(for the keener)
Deep Leaning book by Ian Godfellow, Yoshua Bengio, Aaron Courville: http://www.deeplearningbook.org/contents/rnn.html