Runa Eschenhagen

See [my comment](https://github.com/AlexImmer/Laplace/issues/100#issuecomment-1140018068) in #100.

Clarify in the docs with which models and losses the different BackPACK and ASDL backends are compatible, e.g. that only the ASDL backends can handle `subset_of_weights='all'` with a model not...

Add a list of all constraints on the data loader in the docs, e.g. the batches have to have two elements `X` and `y` and `len(data_loader.dataset)` needs to be implemented.

From #69: We could consider adding more informative error messages when running out of memory during Hessian allocation / computation. E.g., if initialising the Hessian runs out of memory, we...

Check if or which approximations support models with learnable BatchNorm parameters and add clarifying comment to the docs.

Example: continual learning on toy data

2

Simple example showing continual learning with the Laplace approximation on toy data.

A straight forward way to improve the code quality is to enable runtime support for type hints via `typing`.

Users might want to avoid computing the Hessian approximation every time they run their code or reuse the same Laplace approximation in different files (e.g. #42). The best interface would...

Support DataParallel for the predictions and Hessian computation (with Kazuki's backend).

Create a diagram which describes the inheritance structure of all subclasses of `laplace.BaseLaplace`.