This series of paintings is my first use of machine learning as a technology to curate and classify painting compositions.  I use machine learning software:

  • to extract shapes from my existing abstract paintings
  • then to automate creating new compositions
  • then create new color palettes to apply to the shapes
  • then to curate which ones I will like most

This software has created over 40,000 compositions to process. I have been testing different classification processes while developing new ways of processing the paintings including: Linear Regression, Random Forests, Decision Trees, K-Means clusters, and Google TensorFlow. Although I am using machine learning, the most fascinating aspect of the project for me is how much I am learning about my personal subjective preferences and abstract composition.

It has also been interesting documenting the learning process through the act of painting. Painting is seen as a valuable, resource intensive process, while software processing is quick in comparison. I am spending valuable resources painting results of instantaneous algorithms.

The paintings are titled based on Google’s untrained TensorFlow Inception Classification model. “Untrained” means it is downloaded from Google and has not trained for my specific task. The untrained Tensorflow Inception model classifies images to objects, so the software suggest what the abstract painting could be. It is interesting to take a second look at abstract art and consider how its been visually interpreted code. For example the software has classified paintings as hockey puck, bird house, sweet potato, bulletproof vest as some of the objects. If you want to learn more about this process check out this page.


Pin It on Pinterest