Practical resources for machine learning for artists

This tutorial will introduce practical resources, applications, and code for getting started with machine learning for creative and artistic use. It will include an overview of the materials available with ml4a.github.io, as well as a survey of helpful classes, writings, and code made by others in the field.

Most of the session will consist of tutorials for using the Python notebooks and openFrameworks applications bundled with ml4a, including several demonstrations of Wekinator applications. It will also cover the web-based demos of ml4a, including javascript deep learning libraries of interest to web developers. It will end with a broad (but not detailed) survey of various open-source repositories for running deep learning routines to produce images and text (and maybe sound, TBD).

The tutorial is loosely a practical follow-up on the more theoretical lecture "From principal components to puppyslugs" on Tuesday, but attendance to that is not required to take this session.

Tentative list of topics to cover

Data science tasks

• extracting features from images, sounds, text
• search query by feature vectors
• t-SNE and visualization
• training a neural network for image classification
• transfer learning for fine-tuning a neural net to custom image classes
• "neural net painter"

Real-time (mostly using openFrameworks)

• live image classification and object detection
• fast image search and substitution
• train classifier for images and sounds
• controlling music and visual art programs with Wekinator

Web-side demos

• survey of tools used in making ml4a.github.io/demos
• convnet.js, keras.js
• resources for computer vision in the browser