cream cheese

cream cheese explores the effects of applying machine learning and other statistical learning techniques to extremely small amounts of data. The system learns patterns from one image and then tries to recreate a second image while restricted to using the patterns it learned from the first image.

I can use midi controllers to adjust parameters, which causes some of the pixels to cascade and shift. I can use color to capture these changes.

The original texture was from a picture of the eponymous cream cheese, though I usually create custom training images.

More technical details

I use numpy to extract the probabilities. I get to reuse a trick I learned implementing convolutional neural networks.

The system takes inspiration from cellular automata, but I do use randomness (while cellular automata can create seemingly organic random patterns using strict rules) and for me it starts to feel closer to a Markov model.