Making Art like a Neural Network: Optimization of Artistic Practice via Stochastic Gradient Descent

A reappraisal of the artistic process through the lens of A.I./M.L., using the language of backpropagation and optimization algorithms as they relate to deep learning and neural networks. This paper analyzes and discusses the artistic process as an instance of Stochastic Gradient Descent (SGD) performing upon the practice of the artist (a connectionist, non-representational model). This paper elaborates and develops upon the language of such algorithms, including learning rate, momentum, and weight decay as these concepts align with the artistic process, and in so doing offers new insights into the way we think machine learning and neural networks while providing some possible avenues for future artistic research.