# Notes for Meeting April 7 2022
## Single circuit model
* Training with single latent variable
* We're using this SIC-POVM measurements, can we do something better?
* Untrained can be mistaken for extracting EE, but ...
* Training properly it just learns $\theta$
* "properly" means ?
What can we do from here?
* Could be because there is no noise in dataset
* Could be because at single theta there is just one deterministic sample (unlike Ising)
* Could be that we need to add more constraints to loss function (penalize learning input parameter)
* Related to beta-VAE? Other ideas from 'disentangling'