Following David Foster's Generative Deep Learning (2nd. Edition)
Which means I get to learn Julia and Flux as well as how these models perform. This will help me choose tooling for an upcoming project.
I'll be using Flux and since I run mainly on Mac OS: Metal GPU accelerators. I'm looking to develop and benchmark online training of a generative model for a proprietary application, however this project allows me to share some work and results back into the community.
All errors, omissions and bogisities are my own work. I'm interested in application and practical considerations of scale and ergonomics as well as how well the and code/model can be integrated.
Some examples may need to be adapted for the frameworks concerened from the original Tensorflow/Keras in the book cited. I reserve the right to depart from that script and explore avenues more relevant to my misison.
I will also be repeating this exercise in Jax/Flax (Python) so we will have seen GDL2E cooked three ways and hopefully a better understanding of the merits of each framework and language.
I'll be developing the prototype application as I go in Julia as I know Python and I'd like to give Julia a try first.
I won't be using the ubiquitous Jupyter notebook format. All my sources will be in org files using jupyter-emacs for the backends. This means I can mix kernels (languages and environments) willy nilly and write notes so that source code can be tangled and documents extracted. YMMV. See here for how I make this work.
I acknowledge all copyrights of the respective authors of cited works and quotations. Simon Beaumont - England, December 2023 All work here is provided for Research Purposes Only