- The QR decomposition (speed + numerical stability?) and the
R2prior, see: Estimating Regularized Linear Models with rstanarm.
- Priors over correlation matrices from Generating random correlation matrices based on vines and extended onion method. In the single parameter case you can also use beta + bijectors, which is quite cool.
- Gaussian procceses, after attending Principled Priors for Gaussian Processes.
- Find a nice example of a use for the new hyperparameters stuff in TF2, see: this github issue
- Code up some splines from scratch in tensorflow probability, see e.g. splines in stan or try the same methods that jagam uses.
- Look in depth at the tensorflow LinearOperator.