This document covers planned features for this library. More will be added later.
- Implementation of Keras like API for DGPs; for instance, initialising each layer on its own and then stacking the layers as a network.
- Seperatin of model and inference procedure (similar to what GPflow does)
- Ability to create layers with multiple kernels
- Ability to perform multi-task learning
- Doubly Stochastic VI (Salimbeni and Diesenroth 2017)
- Standard VI (Damianou and Lawrence 2013)
- Some for of MCMC