I have previously contributed to PyMC4 and I am willing to implement, test, maintain a higher-level API for Gaussian Processes in PyMC4 using TensorFlow and TensorFlow Probability and also write tutorials and notebooks with expressive animations/figures explaining their usage under GSoC 2020. I also look forward to improve the upstream projects (TensorFlow and TensorFlow Probability) to better support the GP interface of PyMC4.
I have been referring to the following papers to get a deeper understanding of Gaussian Process (but found myself wondering how to implement):
- Introduction to Gaussian Process by David J.C. MacKay
- Gaussian Processes in Machine Learning by Carl Edward Rasmussen
- Gaussian Processes for Machine Learning by Matthias Seeger
Are there any other papers I need to refer from a point of view of implementing them in PyMC4? Do I need to go through the GP API in PyMC3 to implement a similar one in PyMC4 or some other design is desired? I have started working on a proposal but need some direction regarding topics to include and focus on (like project design/timeline). I also wanted to know if it is necessary to read this book cited by PyMC3. I will shortly start working on the project on my fork once I get some direction from the mentors/maintainers.
EDIT: I have created this GitHub Repository containing the projects and my draft project proposal.