Artem Artemev
Artem Artemev
Hello, `haiku` team! Thanks a lot for making awesome `haiku`. I'm interested in sequential probabilistic models. Normally, parameters of probabilistic models are constrained. A simple example would be _variance_. It...
If GPflow were able to handle leading dimensions that would allow running the training of multiple models in parallel. E.g. we could run several models with different initial parameters, maybe...
Refactor expectations code. The recursive nature of the code makes it very difficult to understand, debug and cause maintaining issues. Issues with the code to address: * Flipping types in...
Keras became a core part of TensorFlow (for a good or a bad). The GPflow can benefit from Keras ecosystem. Contact: @awav
TensorFlow probability has multiple features of interest for us: * Distributions. They could replace our custom implementations (Gaussian, GaussianMC) and * Optimisers and samplers. We could hijack some ideas on...
Hello all, I have a very simple question. How to compute gradients w.r.t. a subset of parameters. Let's say I have a model like: ```swift struct Model: Layer { var...
**PR type:** bugfix / enhancement ## Summary GPflow had a feature of setting the minimal positive value using `gpflow.config.set_default_positive_minimum`. Recently, the default minimum boundary for positive values in scalar likelihoods...
It would be nice to have high-performance triangular solve operation implemented (for both lower and upper triangular matrices) in the Triton compiler, similar to the existing `tl.dot`. Triangular solve is...
### Description The subscript adheres to a lower position adjacent to the parentheses surrounding fraction equations, compared to its placement in the LaTeX version. This causes it to hang too...