Overview
The last couple years have seen an explosion in research using AI and machine learning techniques for creative purposes, especially for image creation. If you’ve seen apps like Prisma, or images from Google’s Deep Dream, then you’ve seen these algorithms in action. It’s early days, and while some of the new aesthetics are beautiful, the tools and approaches are still embryonic. In this talk Xavier will show work he and others have made with new algorithms, and explain at a high level how some of these neural-network based tools actually work. In particular he’ll explain his “multiscale neural texture synthesis” algorithm that has allowed for the creation of much more varied and high-resolution images.
Many of these techniques are not yet usable by people without a deep technical background, and Xavier will discuss directions he is taking his work to try to make the algorithms more accessible and expressive. Just as you don’t need to be an audio engineer to create electronic music on a synthesizer, neither should you need to be an AI researcher to create electronic images with a neural net. Just as you don’t expect a synthesizer to write your music, neither should we look to algorithms to create our images.
Objective
To share some of the advances in using AI to create images, and explain how the algorithms work.
Target Audience
Technically curious creatives who want to better understand how visual neural algorithms like style transfer and deep dream actually work.
Assumed Audience Knowledge
None: the talk will be mostly accessible, with just occasional more hardcore diversions
Five Things Audience Members Will Learn
- How a neural network works (at a very high level)
- How style transfer algorithms (like Prisma) work
- Some tools they can use today to create neural images
- Current frontiers in creative AI
- Interesting artists working with AI today
