Daniel Kelshaw

I've also attempted another approach in which I create a child class of `RiemannianMetric`: ```python from geomstats.geometry.riemannian_metric import RiemannianMetric class NumericalMetric(RiemannianMetric): def __init__(self, dim): super().__init__(dim) def metric_matrix(self, base_point): theta =...

Furthermore, your implementation of the closed form `KL Divergence` is the same as seen in *equation 10* of the [Auto-Encoding Variational Bayes](https://arxiv.org/pdf/1312.6114.pdf) paper: https://github.com/JavierAntoran/Bayesian-Neural-Networks/blob/022b9cedb69be3fecc83d4b0efe4b5a848119c2a/src/Bayes_By_Backprop_Local_Reparametrization/model.py#L25-L28 I was wondering if you could...

@JavierAntoran - thank you for taking the time to help explain this, I really appreciate it! I found your explanation of the *local reparameterisation trick* very intuitive and feel like...

Just wondering if there was any update on this? I'd say the `pathlib` library is fairly standard now so it would be nice to have support from `absl` if possible.

@luisfpereira, thanks for getting back. It's been put on hold for a little while, but interested in looking again. Could you provide a brief explanation for how this achieved? If...

@luisfpereira, thank you for the detailed reply. If computing the distance requires the `log` mapping, how can you verify this is the true geodesic distance? There may be several valid...