Objective Functions: Creating Images with Neural Networks
How are advances in artificial intelligence changing aesthetics? What are new kinds of images we can create today, that we could never create before? In this two-part workshop, we will study neural networks, and how algorithms originally designed to allow computers to recognize the objects in photographs have been repurposed to instead create new images.
We'll go over how a particular algorithm architecture works — the convolutional neural network — and the ways in which it is (loosely) inspired by the architecture of the human visual system. We'll discuss how it learns abstracted representations of what exists in an image, and how we can manipulate those abstractions to generate new images.
We will build code to create our own images using these techniques. We'll learn methods that capture the texture of an image: bring your own textured images, for instance of a rug pattern or wood grain, and we'll distil the essence of that texture and then use it to create new abstract images.
The techniques we'll learn are closely related to recent things such as Google's DeepDream, the filters of the Prisma app, or the "painterly" filters built into Facebook Messenger. However, rather than the "push-button" approach of these consumer apps we'll actually explore the algorithmic materiality, and be empowered to understand it and wield it in new ways.
Dates: Saturday, September 23rd & 30th
Annual Studio or Workshop Pass Discount: $150 + $10 Materials Fee
Regular Price: $200 + $10 Materials Fee
What, if any, prerequisite knowledge should students have?
This is an intermediate level course. General computer literacy will be necessary, and experience with programming will be a huge asset, especially in Python. Basic knowledge of linear algebra (matrix multiplication, mostly) will be an asset.
What materials/equipment should students bring?
Students should bring a laptop. Optional: Texture-rich photographs / image files to use as source material. Examples would be, wood grain on a table, a rug with a geometric pattern, or a graffiti covered wall.
About the Instructor:
Xavier Snelgrove is a Toronto-based algorithm designer, entrepreneur, educator and artist. He designs neural networks and other algorithms as CTO at Whirlscape, for apps with AI powered emoji and GIFs. He also organizes the annual GenArtHackParty, that brings both beginners and experienced artists and developers together to learn from each other and build generative artworks—quickly. With a background in computer graphics and computer vision, he has recently been exploring the new aesthetics that neural networks have begun to allow us to create. You can find his work at http://wxs.ca
Questions? If you have any questions about the workshop or special requirements, please contact us at firstname.lastname@example.org. Cancellation and Rescheduling Policy: We are unable to accommodate attendee cancellations or refunds less than 1 week prior to a workshop or event. InterAccess reserves the right to cancel or reschedule this workshop if necessary.