Bayesian active learning for converging posteriors in latent variable inference for control systems | Synapse