Localhost Talk: creative applications of deep learning, aka, neural networks for fun and not profit :-)

Earlier this week I gave a talk at Localhost, the Recurse Center’s public-facing technical speaker series. Slides embedded below. Here’s also a link to the talk slides if you want to see my notes included.

The talk covers some of the creative deep learning projects I’ve worked on while at RC:

• generating jazz with an LSTM [+ github]

• generating punchlines to jokes with seq2seq [+ github]

• generating maps and buildings using circuit boards with pix2pix

• neural style transfer [+ github]

• translating between Pokemon types with CycleGAN [+ dataset] (blog post and repo link to come!)

Overall I received a lot of enthusiastic positive feedback and felt pretty good about how it went! I do feel somewhat proud of all of the fun projects I was able to explore while at RC, and it feels nice to be able to share that with others.

Computational Creativity

I gave a presentation this week about some applications of artificial neural networks in computational creativity. It consists of an overview and discussion of 3 different papers:

1. A Computational Model of Poetic Creativity with Neural Network as Measure of Adaptive Fitness

2. A Neural Algorithm of Artistic Style

3. What Happens Next? Event Prediction Using a Compositional Neural Network Model (part of the What-If Machine project)

Here are the slides: