Fitting GMM through gradient descent

[LemonPi]

Hi Angus, have you ever considered fitting the GMM through a gradient based method instead of EM, such as mentioned here: https://stats.stackexchange.com/questions/64193/fitting-a-gaussian-mixture-model-using-stochastic-gradient-descent ?

I'm interested in a differentiable clustering algorithm for a pipeline where there are some parameters before the clustering to learn. So the clustering itself has to be differentiable to propagate gradients backwards. Do you know a better way to do this?