Consider binarizing the weights in the splat decoder

[quantumjot]

Here, we have a scalar value which adjusts the weight of each Gaussian in the decoder:

https://github.com/alan-turing-institute/affinity-vae/blob/23349e544dc86cf335ab0e948ccea333f40ca3b4/avae/decoders/differentiable.py#L188-L195

We could consider using binarization of this linear layer to improve the reconstruction.

[quantumjot]

Some useful thoughts in this post:
https://discuss.pytorch.org/t/binary-activation-function-with-pytorch/56674/5