Technical Design And Development
For a high level view of how I use machine learning in my artwork I have written Deep Art.
One thing I love about Data Science is that it is a true science which is underpinned with a heavy amount of experimentation and human creativity and endeavour. This is not everyone’s cup of tea and it’s not vital to using the fruits of machine learning. Personally, I enjoy the learning experience of trial and error. I tend to experiment sporadically - mainly with the architecture and the way my machines learn. With that said, I do want to give some idea of the tools I use for my art as they have been fairly consistent in recent years.
In terms of machine learning platforms - I mainly work with Tensorflow but still use Pytorch. I think I am with the consensus, finding the latter better for development and the former better for production ready models. I write the bare minimum of code I need to, but when I do it’s in Python and I tend to use Colab or other Jupyter Notes-like IDEs. I train my models mostly on Google’s cloud platform. My GAN models are derived from Nvidia lab’s original StyleGan work.
I strongly believe that the best AI Art is that which has a high level of human involvement, control and is curated by the choices and guidance of an artist. I personally have a very strong admiration for a several artists and art movements - particularly those of post impressionism and Fauvism. AI Art, like any other human expression has to be anchored in something to avoid being vague and meaningless.
For the field I work in, the seed datasets which I use to train my AI models are key to their subsequent "understanding" of what you wish to achieve as an artist. The datasets I use to create my artwork are entirely my own and private. This is purely to protect the originality and commercial rights of my work.